GDPS/MTMM
In this chapter, we discuss the capabilities and prerequisites of the GDPS/MTMM offering. GDPS/MTMM supports both planned and unplanned situations, helping to maximize application availability and provide business continuity. A GDPS/MTMM solution delivers the following benefits:
•Near-continuous availability
•Disaster recovery (DR) across metropolitan distances
•Recovery time objective (RTO) less than an hour
•Recovery point objective (RPO) of zero
Another key benefit of GDPS/MTMM is that it provides protection against multiple failures. GDPS/MTMM maintains three copies of your data so that even if one copy becomes unavailable, GDPS/MTMM can continue to provide near-continuous availability and DR by using the remaining two copies.
The functions provided by GDPS/MTMM fall into two categories: Protecting your data and controlling the resources managed by GDPS. The following functions are among those that are included:
•Protecting your data:
 – Ensuring the consistency of the secondary copies of your data in the event of a disaster or suspected disaster, including the option to also ensure zero data loss
 – Transparent switching to the secondary disk using HyperSwap
•Controlling the resources managed by GDPS during normal operations, planned changes, and following a disaster:
 – Monitoring and managing the state of the production z/OS systems and LPARs (shutdown, activating, deactivating, IPL, and automated recovery)
 – Monitoring and managing z/VM guests (shutdown, activating, deactivating, IPL, and automated recovery)
 – Managing the couple data sets and coupling facility recovery
 – Support for switching your disk, or systems, or both, to another site
 – User-customizable scripts that control how GDPS/MTMM reacts to specified error situations, which can also be used for planned events
7.1 Introduction to GDPS/MTMM
GDPS/MTMM is a continuous availability and disaster recovery solution that handles many types of planned and unplanned outages. As mentioned in Chapter 1, “Introduction to business resilience and the role of GDPS” on page 1, most outages are planned, and even among unplanned outages, most are not disasters. GDPS/MTMM provides capabilities to help provide the required levels of availability across these outages and in a disaster scenario. These capabilities are described in this chapter.
GDPS/MTMM leverages the IBM MTMM disk mirroring technology to maintain two synchronous secondary copies of your data. The primary copy and each of the two secondary copies are also called disk locations. The three disk locations, or copies, are H1, H2, and H3. H1 and H2 are assumed to be “local” and are fixed in Site 1. H3 is fixed in Site 2.
At any specific point in time, the production systems run on H1, H2 or H3 disk. Whichever copy the production systems are running on is known as the primary disk, and the other two copies are known as the secondary disks. Although the primary disk role can be with any of three disk locations, in a typical configuration:
•The primary disk is in Site1, that is, either H1 or H2.
•The other disk copy in Site1 provides high availability or HA protection.
•The copy in Site 2 (H3) provides disaster recovery or DR protection.
Each of the replication connections between the H1, H2, and H3 locations is called a replication leg or simply a leg. The replication legs in an MTMM configuration have fixed names that are based on the two disk locations that they connect:
•The H1-H2 (or H2-H1) leg is RL1.
•The H1-H3 (or H3-H1) leg is RL2.
•The H2-H3 (or H3-H2) leg is RL3.
The name of a given replication leg never changes, even if the replication direction is reversed for that leg. However, the role of a leg can change, depending on primary disk location. The two legs from the current primary to each of the two secondaries serve as the active replication legs whereas the leg between the two secondary locations serves as the incremental resync or MTIR leg.
To illustrate this concept, consider the sample GDPS/MTMM configuration that is shown in Figure 7-1.
Figure 7-1 Sample GDPS/MTMM Configuration
In this sample configuration, H1 is the primary disk location, RL1 and RL2 are the active replication legs, and RL3 is the MTIR leg.
If there is a disk switch and H2 becomes the new primary disk, RL1 and RL3 become the active replication legs and RL2 becomes the MTIR leg.
7.1.1 Protecting data integrity and data availability with GDPS/MTMM
In 2.2, “Data consistency” on page 17, we point out that data integrity across primary and secondary volumes of data is essential to perform a database restart and accomplish an RTO of less than hour. This section includes details about how GDPS/MTMM automation provides both data consistency if there are mirroring problems and data availability if there are primary disk problems.
Two types of disk problems trigger a GDPS automated reaction:
•PPRC Mirroring problems (Freeze triggers). No problem exists writing to the primary disk subsystem, but a problem exists mirroring the data to one or both of the secondary disk subsystems. For more information, see “GDPS Freeze function for mirroring failures” on page 192.”
•Primary disk problems (HyperSwap triggers). There is a problem writing to the primary disk: either a hard failure, or the disk subsystem is not accessible or not responsive. For more information, see “GDPS HyperSwap function” on page 196.
GDPS Freeze function for mirroring failures
GDPS uses automation, keyed off events or messages, to stop all mirroring for a given replication leg when a remote copy failure occurs between one or more of the primary/secondary disk subsystem pairs on that replication leg. In particular, the GDPS automation uses the IBM PPRC Freeze and Run architecture, which has been implemented as part of Metro Mirror on IBM disk subsystems and also by other enterprise disk vendors. In this way, if the disk hardware supports the Freeze and Run architecture, GDPS can ensure consistency across all data in the sysplex (consistency group), regardless of disk hardware type.
This preferred approach differs from proprietary hardware approaches that work only for one type of disk hardware. For more information about data consistency with synchronous disk mirroring, see “PPRC data consistency” on page 24.
When a mirroring failure occurs, this problem is classified as a Freeze trigger and GDPS stops activity across all disk subsystems for the affected replication leg at the time the initial failure is detected, thus ensuring that the dependant write consistency of the secondary disks for that replication leg is maintained. Note that mirroring activity for the other replication leg is not affected by the freeze.
This is what happens when a GDPS performs a Freeze:
•Remote copy is suspended for all device pairs on the affected replication leg.
•While the suspend command is being processed for each LSS, each device goes into a long busy state. When the suspend completes for each device, z/OS marks the device unit control block (UCB) in all connected operating systems to indicate an Extended Long Busy (ELB) state.
•No I/Os can be issued to the affected devices until the ELB is thawed with the PPRC Run action or until it times out. (The consistency group timer setting commonly defaults to 120 seconds, although for most configurations a longer ELB is preferable.)
•All paths between the PPRCed disks on the affected replication leg are removed, preventing further I/O to the associated secondary disks if PPRC is accidentally restarted.
Because no I/Os are processed for a remote-copied volume during the ELB, dependent write logic ensures the consistency of the affected secondary disks. GDPS performs a Freeze for all LSS pairs that contain GDPS managed mirrored devices.
 
Important: Because of the dependent write logic, it is not necessary for all LSSs to be frozen at the same instant. In a large configuration with many thousands of remote copy pairs, it is not unusual to see short gaps between the times when the Freeze command is issued to each disk subsystem. Because of the ELB, however, such gaps are not a problem.
After GDPS performs the Freeze and the consistency of the secondary disks on the affected leg is protected and the action GDPS takes next depends on the client’s PPRCFAILURE policy (also known as Freeze policy). See “Freeze policy (PPRCFAILURE policy) options” on page 193 for details regarding the actions GDPS will take, based on this policy.
GDPS/MTMM uses a combination of storage subsystem and sysplex triggers to automatically secure, at the first indication of a potential disaster, a data-consistent secondary copy of your data using the Freeze function. In this way, the secondary copy of the data is preserved in a consistent state, perhaps even before production applications are aware of any issues. Ensuring the data consistency of the secondary copy ensures that a normal system restart can be performed instead of having to perform DBMS forward recovery actions. This is an essential design element of GDPS to minimize the time to recover the critical workloads in the event of a disaster in the primary site.
You can appreciate why such a process must be automated. When a device suspends, there is not enough time to launch a manual investigation process. The entire mirror for the affected leg must be frozen by stopping further I/O to it, and then the policy indicates whether production will continue to run with mirroring temporarily suspended, or whether all systems should be stopped to guarantee zero data loss.
In summary, freeze is triggered as a result of a PPRC suspension event for any primary disk in the GDPS configuration, that is, at the first sign that a duplex mirror that is going out of the duplex state. When a device suspends, all attached systems are sent a “State Change Interrupt” (SCI). A message is issued in all of those systems and then each system must issue multiple I/Os to investigate the reason for the suspension event.
When GDPS performs a freeze, all primary devices in the PPRC configuration suspend for the affected replication leg. This can result in significant SCI traffic and many messages in all of the systems. GDPS, in conjunction with z/OS and microcode on the DS8000 disk subsystems, supports reporting suspensions in a summary message per LSS instead of at the individual device level. When compared to reporting suspensions on a per devices basis, the Summary Event Notification for PPRC Suspends (PPRCSUM) dramatically reduces the message traffic and extraneous processing associated with PPRC suspension events and freeze processing.
Freeze policy (PPRCFAILURE policy) options
As we have described, when a mirroring failure is detected on a replication leg, GDPS automatically and unconditionally performs a Freeze of that leg to secure a consistent set of secondary volumes in case the mirroring failure could be the first indication of a site failure. Because the primary disks are in the Extended Long Busy state as a result of the freeze and the production systems are locked out, GDPS must take some action. Here, there is no time to interact with the operator on an event-by-event basis. The action must be taken immediately. The action to be taken is determined by a customer policy setting, that is, the PPRCFAILURE policy option (also known as the Freeze policy option). GDPS will use this same policy setting after every Freeze event to determine what its next action should be. The policy can be specified at a leg level allowing a different policy specification for the replication legs. The options are as follows:
•PPRCFAILURE=GO (Freeze and Go)
GDPS allows production systems to continue operation after mirroring is suspended.
•PPRCFAILURE=STOP (Freeze and Stop)
GDPS resets production systems while I/O is suspended.
•PPRCFAILURE=STOPLAST
GDPS checks the mirroring status of the other replication leg. If the status of the other leg is OK, GDPS performs a Go. If not, and this is the last viable leg that GDPS has just frozen, GDPS performs a Stop.
•PPRCFAILURE=COND (Freeze and Stop conditionally)
GDPS tries to determine if a secondary disk caused the mirroring failure. If so, GDPS performs a Go. If not, GDPS performs a Stop.
•PPRCFAILURE=CONDLAST
GDPS checks the mirroring status of the other replication leg. If the status of the other leg is OK, GDPS performs a Go. If not (the freeze was performed on the last viable leg), GDPS tries to determine if a secondary disk caused the mirroring failure. If so, GDPS performs a Go. If not, GDPS performs a Stop.
Freeze and Go
With this policy, after performing the Freeze, GDPS performs a Run action against all primary LSSs, which is also known as performing a Go. Performing a Go removes the ELB and allows production systems to continue using these devices. The devices will be in remote copy-suspended mode in relation to the secondary devices on the affected leg, so any further writes to these devices are no longer being mirrored to the secondary devices on that leg. However, changes are being tracked by the hardware so that, later, only the changed data will be resynchronized to the affected secondary disks.
With this policy you avoid an unnecessary outage for a false freeze event, that is, if the trigger is simply a transient event. However, if the trigger turns out to be the first sign of an actual disaster, you might continue operating for an amount of time before all systems fail. Any updates made to the primary volumes during this time are not replicated to the secondary disk, and therefore are lost if you end up having to recover on the affected secondary disk. In addition, because the CF structures were updated after the secondary disks were frozen, the CF structure content is not consistent with the secondary disks. Therefore, the CF structures in either site cannot be used to restart workloads and log-based restart must be used when restarting applications.
This is not full forward recovery. It is forward recovery of any data, such as DB2 group buffer pools, that might have existed in a CF but might not have been written to disk yet. This results in prolonged recovery times. The duration depends on how much such data existed in the CFs at that time. With a Freeze and Go policy, you might consider tuning applications such as DB2, which can harden such data on disk more frequently than otherwise.
Freeze and Go is a high availability option that avoids production outage for false freeze events. However, it carries a potential for data loss.
Freeze and Stop
With this policy, you can be assured that no updates are made to the primary volumes after the Freeze because all systems that can update the primary volumes are reset. This ensures that no more updates can occur to the primary disks because such updates would not be mirrored to the affected secondary disk, meaning that it would not be possible to achieve zero data loss if a failure occurs (or if the original trigger was an indication of a catastrophic failure) and recovery on the affected secondary disk is required.
You can choose to restart the systems when you want. For example, if this was a false freeze (that is, a false alarm), then you can quickly resynchronize the mirror and restart the systems only after the mirror is duplex.
If you are using duplexed coupling facility (CF) structures along with a Freeze and Stop policy, it might seem that you are guaranteed to use the duplexed instance of your structures if you must recover and restart your workload with the frozen secondary copy of your disks. However, this is not always the case. There can be rolling disaster scenarios where before, after, or during the freeze event, there is an interruption (perhaps failure of CF duplexing links) that forces CFRM to drop out of duplexing. There is no guarantee that it is the structure instance in the surviving site that is kept. It is possible that CFRM keeps the instance in the site that is about to totally fail. In this case, there will not be an instance of the structure in the site that survives the failure.
To summarize, with a Freeze and Stop policy, if there is a surviving, accessible instance of application-related CF structures, this instance will be consistent with the frozen secondary disks. However, depending on the circumstances of the failure, even with structures duplexed across two sites you are not 100% guaranteed to have a surviving, accessible instance of the application structures and therefore you must have the procedures in place to restart your workloads without the structures.
Although a Stop policy can be used to ensure no data loss, if a failure occurs that is a false freeze event, that is, it is a transient failure that did not necessitate recovering using the frozen disks, it results in unnecessarily stopping the systems.
Freeze and Stop last
With this policy, after the Freeze, GDPS checks the status of mirroring on the other replication leg (the leg other than the one that was just frozen) to determine whether the leg that just frozen was the last leg actively replicating data. If the other leg is still actively replicating data, GDPS performs a Go. But if the other leg is already frozen or mirroring status is not OK, GDPS performs a Stop.
When you have only one replication leg defined in your configuration (you have only one secondary copy of your data), using this policy specification is the same as using a Freeze and Stop policy.
Freeze and Stop conditional
Field experience has shown that most of the Freeze triggers are not necessarily the start of a rolling disaster, but are “False Freeze” events that do not necessitate recovery on the secondary disk. Examples of such events include connectivity problems to the secondary disks and secondary disk subsystem failure conditions.
With a COND policy, the action that GDPS takes after it performs the Freeze is conditional. GDPS tries to determine if the mirroring problem was as a result of a permanent or temporary secondary disk subsystem problem:
•If GDPS can determine that the freeze was triggered as a result of a secondary disk subsystem problem, GDPS performs a Go. That is, it allows production systems to continue to run by using the primary disks. However, updates will not be mirrored until the secondary disk can be fixed and PPRC can be resynchronized.
•If GDPS cannot ascertain that the cause of the freeze was a secondary disk subsystem, GDPS operates on the assumption that this could still be the beginning of a rolling disaster in the primary site and performs a Stop, resetting all the production systems to guarantee zero data loss. GDPS cannot always detect that a particular freeze trigger was caused by a secondary disk, and that some freeze events that are in fact caused by a secondary disk could still result in a Stop.
For GDPS to determine whether a freeze trigger might have been caused by the secondary disk subsystem, the IBM DS8000 disk subsystems provide a special query capability known as the Query Storage Controller Status microcode function. If all disk subsystems in the GDPS managed configuration support this feature, GDPS uses this special function to query the secondary disk subsystems in the configuration to understand the state of the secondaries and if one of these secondaries might have caused the freeze. If you use the COND policy setting but all disks in your configuration do not support this function, GDPS cannot query the secondary disk subsystems, and the resulting action is a Stop.
This option can provide a good compromise where you can minimize the chance that systems would be stopped for a false freeze event and increase the chance of achieving zero data loss for a real disaster event.
Freeze and Stop conditional last
With this policy, after the Freeze, GDPS checks the status of mirroring on the other replication leg (the leg other than the one that was just frozen) to determine if the leg just frozen was the last leg actively replicating data. If the other leg is still actively replicating data, GDPS performs a Go. If the other leg is already frozen or mirroring status is not OK, GDPS performs conditional Stop processing; that is, it queries the secondary disk subsystem and performs a Go if, as a result of the query, it determines that the freeze was caused by the secondary, but performs a Stop if it cannot determine for sure that the problem was caused by the secondary.
When you only have one replication leg defined in your configuration (you only have one secondary copy of your data), using this policy specification is the same as using a Freeze and Stop conditional policy.
PPRCFAILURE policy selection considerations
The PPRCFAILURE policy option specification directly relates to recovery time and recovery point objectives (RTO and RPO, respectively), which are business objectives.Therefore, the policy option selection is really a business decision rather than an IT decision. If data associated with your transactions is high-value, it might be more important to ensure that no data associated with your transactions is ever lost, so you might decide on a Freeze and Stop policy. If you have huge volumes of relatively low-value transactions, you might be willing to risk some lost data in return for avoiding unnecessary outages with a Freeze and Go policy. The Freeze and Stop Conditional policy attempts to minimize the chance of unnecessary outages and the chance of data loss; however, there is still a risk of either, however small.
The various PPRCFAILURE policy options, combined with the fact that the policy options are specified on a per replication leg basis (different policies can be specified for different legs), gives you the flexibility to refine your policies to meet your unique business goals.
For example, if your RPO is zero, you can employ the following PPRCFAILURE policy:
•For RL2, Freeze and Stop (PPRCFAILURE=STOP)
Since H3 is your disaster recovery copy and you must ensure that you never lose data should you ever have to recover and run on the H3 disk, you must always, unconditionally stop the systems to ensure that no further updates occur to the primary disks that could be lost in a recovery scenario.
•For RL1, Freeze and Stop on last leg only (STOPLAST)
You do not need to take a production outage when PPRC freezes on the high-availability leg if RL2 is still functional and continues to provide disaster recovery protection. However, if RL2 is not functional when PPRC on RL1 suspends, you might want to at least retain the capability to recover on H2 disk with zero data loss if it becomes necessary.
However, if you want to avoid unnecessary outages at the risk of losing data if there is an actual disaster, you can specify Freeze and Go for both of your replication legs.
GDPS HyperSwap function
If there is a problem writing or accessing the primary disk because of a failing, failed, or inaccessible primarynon-responsive disk, there is a need to swap from the primary disks to one of the sets of secondary disks.
GDPS/MTMM delivers a powerful function known as HyperSwap. HyperSwap provides the ability to swap from using the primary devices in a mirrored configuration to using what had been one of the sets of secondary devices, in a manner that is transparent to the production systems and applications using these devices. Before the availability of HyperSwap, a transparent disk swap was not possible. All systems using the primary disk would have been shut down (or might have failed, depending on the nature and scope of the failure) and would have been re-IPLed using the secondary disks. Disk failures were often a single point of failure for the entire sysplex.
With HyperSwap, such a switch can be accomplished without IPL and with just a brief hold on application I/O. The HyperSwap function is completely controlled by automation, thus allowing all aspects of the disk configuration switch to be controlled through GDPS.
HyperSwap can be invoked in two ways:
•Planned HyperSwap
A planned HyperSwap is invoked by operator action using GDPS facilities. One example of a planned HyperSwap is where a HyperSwap is initiated in advance of planned disruptive maintenance to a disk subsystem.
•Unplanned HyperSwap
An unplanned HyperSwap is invoked automatically by GDPS, triggered by events that indicate the primary disk problem.
Primary disk problems can be detected as a direct result of an I/O operation to a specific device that fails because of a reason that indicates a primary disk problem such as:
 – No paths available to the device
 – Permanent error
 – I/O timeout
In addition to a disk problem being detected as a result of an I/O operation, it is also possible for a primary disk subsystem to proactively report that it is experiencing an acute problem. The IBM DS8000 provides a special microcode function known as the Storage Controller Health Message Alert capability. Problems of different severity are reported by disk subsystems that support this capability. Those problems classified as acute are also treated as HyperSwap triggers. After systems are swapped to use the secondary disks, the disk subsystem and operating system can try to perform recovery actions on the former primary without impacting applications since the applications are no longer using those disks.
Planned and unplanned HyperSwap have requirements in terms of the physical configuration, such as having to be symmetrically configured, and so on. While a client’s environment meets these requirements, there is no special enablement required to perform planned swaps. Unplanned swaps are not enabled by default and must be enabled explicitly as a policy option. This is described in more detail in “Preferred Swap Leg and HyperSwap (Primary Failure) policy options” on page 199.
When a swap is initiated, GDPS always validates various conditions to ensure that it is safe to swap. For example, if the mirror is not fully duplex on a given leg, that is, not all volume pairs are in a duplex state, a swap cannot be performed on that leg. The way that GDPS reacts to such conditions changes depending on the condition detected and whether the swap is a planned or unplanned swap.
Assuming that there are no show-stoppers and the swap proceeds, for both planned and unplanned HyperSwap, the systems that are using the primary volumes will experience a temporary pause in I/O processing. GDPS blocks I/O both at the channel subsystem level by performing a Freeze which results in all disks going into Extended Long Busy, and also in all systems, where I/O is quiesced at the operating system (UCB) level. This is to ensure that no systems use the disks until the switch is complete. During the time when I/O is paused, the following process is completed:
1. The PPRC configuration is physically switched. This includes physically changing the secondary disk status to primary. Secondary disks are protected and cannot be used by applications. Changing their status to primary allows them to come online to systems and be used.
2. The disks will be logically switched in each of the systems in the GDPS configuration. This involves switching the internal pointers in the operating system control blocks (UCBs). After the switch, the operating system will point to the former secondary devices which will be the new primary devices.
3. Finally, the systems resume operation using the new, swapped-to primary devices. The applications are not aware of the fact that different devices are now being used.
This brief pause during which systems are locked out of performing I/O is known as the User Impact Time. In benchmark measurements at IBM using currently supported releases of GDPS and IBM DS8000 disk subsystems, the User Impact Time to swap 10,000 pairs across 16 systems during an unplanned HyperSwap was less than 10 seconds. Most implementations are actually much smaller than this and typical impact times in a well-configured environment using the most current storage and server hardware are measured in seconds. Although results will depend on your configuration, these numbers give you a high-level idea of what to expect.
HyperSwap can be executed on either replication leg in a GDPS/MTMM environment. For a planned swap, you must specify which leg you want to use for the swap. For an unplanned swap, which leg is chosen depends on many factors, including your HyperSwap policy. This is described in more detail in “Preferred Swap Leg and HyperSwap (Primary Failure) policy options” on page 199.
After a replication leg is selected for the HyperSwap, GDPS swaps all devices on the selected replication leg. Just as the Freeze function applies to the entire consistency group, HyperSwap is for the entire consistency group. For example, if a single mirrored volume fails and HyperSwap is invoked, processing is swapped to one of the sets of secondary devices for all primary volumes in the configuration, including those in other, unaffected, disk subsystems. This is to ensure that all primary volumes remain in the same site. If HyperSwap were to swap only the failed LSS, you would then have several primaries in one location, and the remainder in another location. This would make for a significantly complex environment to operate and administer I/O configurations.
Incremental Resynchronization
When a disk switch or recovery on one of the secondaries occurs, MTMM provides for a capability known as “incremental resynchronization” (IR). Assume your H1 disks are the current primaries and the H2 and H3 disks are the current secondaries. If you switch from using H1 to using H2 as your primary disks, to maintain a multi-target configuration, you will need to establish replication on RL1, between H2 and H1, and on RL3, between H2 and H3. A feature of the PPRC copy technology known as Failover/Failback, together with the MTMM IR capability allows you to establish replication for RL1 and RL3 without having to copy all of the data from H2 to H1 or from H2 to H3. Only the changes that occur on B after primary is switched to B are copied in order to resynchronize the two legs.
If there is an unplanned HyperSwap from H1 to H2, because H1 has failed, replication can be established on RL3 between H2 and H3 in order to restore disaster recovery readiness. Again, this is an incremental resynchronization (only changed tracks are copied), so the duration to get to a protected position will be much faster compared to performing an initial copy for the leg.
HyperSwap with less than full channel bandwidth
You may consider enabling unplanned HyperSwap on the cross-site replication leg (RL2), even if you do not have sufficient cross-site channel bandwidth to sustain the full production workload for normal operations. Assuming that a disk failure is likely to cause an outage and that you have to switch to using the H3 disk in the other site (because the H2 disks in the same site are down at the time), the unplanned HyperSwap to H3 might at least present you with the opportunity to perform an orderly shutdown of your systems first. Shutting down your systems cleanly avoids the complications and restart time elongation associated with a crash-restart of application subsystems.
Preferred Swap Leg and HyperSwap (Primary Failure) policy options
Clients might prefer not to immediately enable their environment for unplanned HyperSwap when they first implement GDPS. For this reason, unplanned HyperSwap is not enabled by default. However, we strongly suggest that all GDPS/MTMM clients enable their environment for unplanned HyperSwap, at a minimum, on the local replication leg (RL1). Both copies of disk on the RL1 leg (H1 and H2) are local and therefore distance and connectivity should not be an issue.
You control the actions that GDPS takes for primary disk problems by specifying a Primary Failure policy option. This option is applicable to both replication legs. However, you have the option of overriding this specification at a leg level and request a different action based on which leg is selected by GDPS to act upon. Furthermore, there is the Preferred Swap Leg policy, which is factored in when GDPS decides which leg to act upon as a result of a primary disk problem trigger.
Preferred Swap Leg selection for unplanned HyperSwap
A primary disk problem trigger is common to both replication legs since the primary disk is common to both legs. Before acting on the trigger, GDPS first needs to select which leg to act upon. GDPS provides you with the ability to influence this decision by specifying a Preferred Swap Leg policy. GDPS will attempt to select the leg that you have identified as the Preferred Swap Leg first. However, if this leg is not eligible for the action that you specified in your Primary Failure policy, GDPS attempts to select the other active replication leg. These are among the reasons that your Preferred Swap Leg might not be eligible for selection:
•It is currently the MTIR leg.
•All pairs for the leg are not in a duplex state.
•It is currently not HyperSwap enabled.
HyperSwap retry on non-preferred leg
If the preferred leg is viable and selected for an unplanned swap, there is still a possibility (albeit small) that the swap on this leg fails for some reason. When swap on the first leg fails, if the other replication leg is enabled for HyperSwap, GDPS will retry the swap on the other leg. This maximizes the chances of a successful swap.
Primary failure policy options
After GDPS has selected which leg it will act on when a primary disk problem trigger occurs, the first thing it will do will be a Freeze on the selected leg (the same as is performed when a mirroring problem trigger is encountered). GDPS then applies the Primary Failure policy option specified for that leg. The Primary Failure policy for each leg can specify a different action. You can specify the following Primary Failure policy options:
•PRIMARYFAILURE=GO
No swap is performed. The action GDPS takes is the same as for a freeze event with policy option PPRCFAILURE=GO. A Run action is performed, which will allow systems to continue using the original primary disks. PPRC is suspended and therefore updates are not being replicated to the secondary. Note, however, that depending on the scope of the primary disk problem, it might be that some or all production workloads simply cannot run or cannot sustain required service levels. Such a situation might necessitate restarting the systems on the secondary disks. Because of the freeze, the secondary disks are in a consistent state and can be used for restart. However, any transactions that ran after the Go action will be lost.
•PRIMARYFAILURE=STOP
No swap is performed. The action GDPS takes is the same as for a freeze event with policy option PPRCFAILURE=STOP. GDPS system-resets all the production systems. This ensures that no further I/O occurs. After performing situation analysis, if it is determined that this was not a transient issue and that the secondaries should be used to IPL the systems again, no data will be lost.
•PRIMARYFAILURE=SWAP,swap_disabled_action
The first parameter, SWAP, indicates that after performing the Freeze, GDPS will proceed with performing an unplanned HyperSwap. When the swap is complete, the systems will be running on the new, swapped-to primary disks (former secondaries). Mirroring on the selected leg will be in a suspended state; because the primary disks are known to be in a problematic state, there is no attempt to reverse mirroring. After the problem with the primary disks is fixed, you can instruct GDPS to resynchronize PPRC from the current primaries to the former ones (which are now considered to be secondaries).
The second part of this policy, swap_disabled_action, indicates what GDPS should do if HyperSwap had been temporarily disabled by operator action at the time the trigger was encountered. Effectively, an operator action has instructed GDPS not to perform a HyperSwap, even if there is a swap trigger. GDPS has already performed a freeze. The second part of the policy control what action GDPS will take next.
The following options (which are in effect only if HyperSwap is disabled by the operator) are available for the second parameter (remember that the disk is already frozen):
GO This is the same action as GDPS would have performed if the policy option had been specified as PRIMARYFAILURE=GO.
STOP This is the same action as GDPS would have performed if the policy option had been specified as PRIMARYFAILURE=STOP.
Preferred Swap Leg and Primary Failure policy selection considerations
For the Preferred Swap Leg policy, consider whether you can tolerate running with disk and systems in opposite sites with no/minimal performance impact. If that is acceptable, you can choose either leg, although it might be better to prefer the RL2 (Site1-Site2) leg. If you cannot tolerate running with disks and systems in opposite sites, choose the RL1, local leg.
For the Primary Failure policy, again we recommend that you specify SWAP for the first part of the policy option to enable HyperSwap, at least on the local replication leg (RL1). If distance and connectivity between your sites is not an issue, consider specifying SWAP for the first part of the policy on the remote replication leg (RL2) also.
For the Stop or Go choice, either as the second part of the policy option or if you will not be using SWAP, similar considerations apply as for the PPRCFAILURE policy options to Stop or Go. Go carries the risk of data loss if it is necessary to abandon the primary disk and restart systems on the secondary. Stop carries the risk of taking an unnecessary outage if the problem was transient. The key difference is that with a mirroring failure, the primary disks are not broken. When you allow the systems to continue to run on the primary disk with the Go option, other than a disaster (which is low probability), the systems are likely to run with no problems. With a primary disk problem, with the Go option, you are allowing the systems to continue running on what are known to be disks that experienced a problem just seconds ago. If this was a serious problem with widespread impact, such as an entire disk subsystem failure, the applications will experience severe problems. Some transactions might continue to commit data to those disks that are not broken. Other transactions might be failing or experiencing serious service time issues. Also, if there is a decision to restart systems on the secondary because the primary disks are simply not able to support the workloads, there will be data loss. The probability that a primary disk problem is a real problem that will necessitate restart on the secondary disks is much higher when compared to a mirroring problem. A Go specification in the Primary Failure policy increases your risk of data loss.
If the primary failure was of a transient nature, a Stop specification results in an unnecessary outage. However, with primary disk problems, the probability that the problem could necessitate restart on the secondary disks is high, so a Stop specification in the Primary Failure policy avoids data loss and facilitates faster restart.
The considerations relating to CF structures with a PRIMARYFAILURE event are similar to a PPRCFAILURE event. If there is an actual swap, the systems continue to run and continue to use the same structures as they did before the swap; the swap is transparent. With a Go action, because you continue to update the CF structures along with the primary disks after the Go, if you need to abandon the primary disks and restart on the secondary, the structures are inconsistent with the secondary disks and are not usable for restart purposes. This will prolong the restart, and therefore your recovery time. With Stop, if you decide to restart the systems using the secondary disks, there is no consistency issue with the CF structures because no further updates occurred on either set of disks after the trigger was captured.
GDPS use of DS8000 functions
GDPS strives to use (when it makes sense) enhancements to the IBM DS8000 disk technologies. In this section we provide information about the key DS8000 technologies that GDPS supports and uses.
PPRC Failover/Failback support
When a primary disk failure occurs and the disks are switched to the secondary devices, PPRC Failover/Failback (FO/FB) support eliminates the need to do a full copy when reestablishing replication in the opposite direction. Because the primary and secondary volumes are often in the same state when the freeze occurred, the only differences between the volumes are the updates that occur to the secondary devices after the switch.
Failover processing sets the secondary devices to primary suspended status and starts change recording for any subsequent changes made. When the mirror is reestablished with failback processing, the original primary devices become secondary devices and a resynchronization of changed tracks takes place.
GDPS/MTMM requires PPRC FO/FB capability to be available on all disk subsystems in the managed configuration.
PPRC eXtended Distance (PPRC-XD)
PPRC-XD (also known as Global Copy) is an asynchronous form of the PPRC copy technology. GDPS uses PPRC-XD rather than synchronous PPRC to reduce the performance impact of certain remote copy operations that potentially involve a large amount of data. See 7.6.2, “Reduced impact initial copy and resynchronization” on page 225 for details.
Storage Controller Health Message Alert
This facilitates triggering an unplanned HyperSwap proactively when the disk subsystem reports an acute problem that requires extended recovery time. See “GDPS HyperSwap function” on page 196 for more information about unplanned HyperSwap triggers.
PPRC Summary Event Messages
GDPS supports the DS8000 PPRC Summary Event Messages (PPRCSUM) function which is aimed at reducing the message traffic and the processing of these messages for Freeze events. This is described in “GDPS Freeze function for mirroring failures” on page 192.
Soft Fence
Soft Fence provides the capability to block access to selected devices. As discussed in “Protecting secondary disks from accidental update” on page 203, GDPS uses Soft Fence to avoid write activity on disks that are exposed to accidental update in certain scenarios.
On-demand dump (also known as non-disruptive statesave)
When problems occur with disk subsystems such as those which result in an unplanned HyperSwap, a mirroring suspension or performance issues, a lack of diagnostic data from the time the event occurs can result in difficulties in identifying the root cause of the problem. Taking a full statesave can lead to temporary disruption to host I/O and is often frowned upon by clients for this reason. The on-demand dump (ODD) capability of the disk subsystem facilitates taking a non-disruptive statesave (NDSS) at the time that such an event occurs. The microcode does this automatically for certain events such as taking a dump of the primary disk subsystem that triggers a PPRC freeze event and also allows an NDSS to be requested by an exploiter. This enables first failure data capture (FFDC) and thus ensures that diagnostic data is available to aid problem determination. Be aware that not all information that is contained in a full statesave is contained in an NDSS and therefore there may still be failure situations where a full statesave is requested by the support organization.
GDPS provides support for taking an NDSS using the remote copy panels. In addition to this support, GDPS autonomically takes an NDSS if there is an unplanned Freeze or HyperSwap event.
Query Host Access function
When a PPRC disk pair is being established, the device that is the target (secondary) must not be used by any system. The same is true when establishing a FlashCopy relationship to a target device. If the target is in use, the establishment of the PPRC or FlashCopy relationship fails. When such failures occur, it can be a tedious task to identify which system is holding up the operation.
The Query Host Access disk function provides the means to query and identify what system is using a selected device. GDPS uses this capability and adds usability in several ways:
•Query Host Access identifies the LPAR that is using the selected device through the CPC serial number and LPAR number. It is still a tedious job for operations staff to translate this information to a system or CPC and LPAR name. GDPS does this translation and presents the operator with more readily usable information, thereby avoiding this additional translation effort.
•Whenever GDPS is requested to perform a PPRC or FlashCopy establish operation, GDPS first performs Query Host Access to see if the operation is expected to succeed or fail as a result of one or more target devices being in use. GDPS alerts the operator if the operation is expected to fail, and identifies the target devices in use and the LPARs holding them.
•GDPS continually monitors the target devices defined in the GDPS configuration and alerts operations to the fact that target devices are in use when they should not be. This allows operations to fix the reported problems in a timely manner.
•GDPS provides the ability for the operator to perform ad hoc Query Host Access to any selected device using the GDPS panels.
Protecting secondary disks from accidental update
A system cannot be IPLed using a disk that is physically a PPRC secondary disk because PPRC secondary disks cannot be brought online to any systems. However, a disk can be secondary from a GDPS (and application use) perspective but physically, from a PPRC perspective, have simplex or primary status.
For both planned and unplanned HyperSwap, and a disk recovery, GDPS changes former secondary disks to primary or simplex state. However, these actions do not modify the state of the former primary devices, which remain in the primary state. Therefore, the former primary devices remain accessible and usable even though they are considered to be the secondary disks from a GDPS perspective. This makes it is possible to accidentally update or IPL from the wrong set of disks. Accidentally using the wrong set of disks can potentially result in a loss of data integrity or data.
GDPS/MTMM provides protection against using the wrong set of disks in different ways:
•If you attempt to load a system through GDPS (either script or panel) using the wrong set of disks, GDPS rejects the load operation.
•If you used the HMC rather than GDPS facilities for the load, then early in the IPL process, during initialization of GDPS, if GDPS detects that the system coming up has just been IPLed using the wrong set of disks, GDPS will quiesce that system, preventing any data integrity problems that could be experienced had the applications been started.
•GDPS uses a DS8000 disk subsystem capability, which is called Soft Fence for configurations where the disks support this function. Soft Fence provides the means to fence (that is, block) access to a selected device. GDPS uses Soft Fence when appropriate to fence devices that would otherwise be exposed to accidental update.
7.1.2 Protecting other CKD data
Systems that are fully managed by GDPS are known as GDPS managed systems or GDPS systems. There are two types of GDPS Systems as follows:
•z/OS systems in the GDPS sysplex
•z/VM systems managed by GDPS/MTMM MultiPlatform Resiliency for System z (xDR)
GDPS/MTMM can also manage the disk mirroring of CKD disks used by systems outside of the sysplex: other z/OS systems, Linux on System z, VM, and VSE systems that are not running any GDPS/MTMM or xDR automation. These are known as “foreign systems.”
Because GDPS manages PPRC for the disks used by these systems, these disks will be attached to the GDPS controlling systems. With this setup, GDPS is able to capture mirroring problems and will perform a freeze. All GDPS managed disks belonging to the GDPS systems and these foreign systems are frozen together, regardless of whether the mirroring problem is encountered on the GDPS systems’ disks or the foreign systems’ disks.
GDPS/MTMM is not able to directly communicate with these foreign systems. For this reason, GDPS automation will not be aware of certain other conditions such as a primary disk problem that is detected by these systems. Because GDPS will not be aware of such conditions that would have otherwise driven autonomic actions such as HyperSwap, GDPS will not react to these events.
If an unplanned HyperSwap occurs (because it was triggered on a GDPS managed system), the foreign systems cannot and will not swap to using the secondaries. A setup is prescribed to set a long Extended Long Busy time-out for these systems such that when the GDPS managed systems swap, these systems hang. The ELB prevents these systems from continuing to use the former primary devices. You can then use GDPS automation facilities to reset these systems and re-IPL them using the swapped-to primary disks.
7.2 GDPS/MTMM configurations
At its most basic, a GDPS/MTMM configuration consists of at least one production system, at least one controlling system in a sysplex, primary disks, and secondary disks. The actual configuration depends on your business and availability requirements. The following three configurations are most common:
•Single-site workload configuration
In this configuration, all of the production systems normally run in the same site, referred to as Site1, and the GDPS controlling system runs in Site2. In effect, Site1 is the active site for all production systems. The controlling system in Site2 is running and resources are available to move production to Site2, if necessary, for a planned or unplanned outage of Site1. Although you might also hear this referred to as an Active/Standby GDPS/MTMM configuration, we avoid the Active/Standby term to avoid confusion with the same term used in conjunction with the GDPS/Active-Active product.
•Multisite workload configuration
In this configuration, the production systems run in both sites, Site1 and Site2. This configuration typically uses the full benefits of data sharing available with a Parallel Sysplex. Having two GDPS controlling systems, one in each site, is preferable. Although you might also hear this referred to as an Active/Active GDPS/MTMM configuration, we avoid the Active/Active term to avoid confusion with the same term used in conjunction with the GDPS/Active-Active product.
•Business Recovery Services (BRS) configuration
In this configuration, the production systems and the controlling system are all in the same site, referred to as Site1. Site2 can be a client site or can be owned by a third-party recovery services provider (thus the name BRS). You might hear this referred to as an Active/Cold configuration.
These configuration options are described in more detail in the following sections.
7.2.1 Controlling system
Why does a GDPS/MTMM configuration need a controlling system? At first, you might think this is an additional infrastructure overhead. However, when you have an unplanned outage that affects production systems or the disk subsystems, it is crucial to have a system such as the controlling system that can survive failures that might have impacted other portions of your infrastructure. The controlling system allows you to perform situation analysis after the unplanned event to determine the status of the production systems or the disks, and then to drive automated recovery actions. The controlling system plays a vital role in a GDPS/MTMM configuration.
The controlling system must be in the same sysplex as the production system (or systems) so it can see all the messages from those systems and communicate with those systems. However, it shares an absolute minimum number of resources with the production systems (typically just the couple data sets). By being configured to be as self-contained as possible, the controlling system will be unaffected by errors that can stop the production systems (for example, an Extended Long Busy event on a primary volume).
The controlling system must have connectivity to all the Site1 and Site2 primary and secondary devices that it will manage. If available, it is preferable to isolate the controlling system infrastructure on a disk subsystem that is not housing mirrored disks that are managed by GDPS.
The controlling system is responsible for carrying out all recovery actions following a disaster or potential disaster, for managing the disk mirroring configuration, for initiating a HyperSwap, for initiating a freeze and implementing the freeze/swap policy actions, for reassigning STP roles; for re-IPLing failed systems, and so on.
 
Note: The availability of the dedicated GDPS controlling system (or systems) in all configurations is a fundamental requirement of GDPS. It is not possible to merge the function of the controlling system with any other system that accesses or uses the primary volumes or other production resources.
Configuring GDPS/MTMM with two controlling systems, one in each site is highly recommended. This is because a controlling system is designed to survive a failure in the opposite site of where the primary disks are. Primary disks are normally in Site1 and the controlling system in Site2 is designed to survive if Site1 or the disks in Site1 fail. However, if you reverse the configuration so that primary disks are now in Site2, the controlling system is in the same site as the primary disks. It will certainly not survive a failure in Site2 and might not survive a failure of the disks in Site2 depending on the configuration. Configuring a controlling system in both sites ensures the same level of protection, no matter which site is the primary disk site. When two controlling systems are available, GDPS manages assigning a Master role to the controlling system that is in the same site as the secondary disks and switching the Master role if there is a disk switch.
Improved controlling system availability: Enhanced timer support
Normally, a loss of synchronization with the sysplex timing source will generate a disabled console WTOR that suspends all processing on the LPAR, until a response is made to the WTOR. The WTOR message is IEA394A in STP timing mode.
In a GDPS environment, z/OS is aware that a given system is a GDPS controlling system and will allow a GDPS controlling system to continue processing even when the server it is running on loses its time source and becomes unsynchronized. The controlling system is therefore able to complete any freeze or HyperSwap processing it might have started and is available for situation analysis and other recovery actions, instead of being in a disabled WTOR state.
In addition, because the controlling system is operational, it can be used to help in problem determination and situation analysis during the outage, thus further reducing the recovery time needed to restart applications.
The controlling system is required to perform GDPS automation in the event of a failure. Actions might include these tasks:
•Reassigning STP roles
•Performing the freeze processing to guarantee secondary data consistency
•Coordinating HyperSwap processing
•Executing a takeover script
•Aiding with situation analysis
Because the controlling system needs to run with only a degree of time synchronization that allows it to correctly participate in heartbeat processing with respect to the other systems in the sysplex, this system should be able to run unsynchronized for a period of time (80 minutes) using the local time-of-day (TOD) clock of the server (referred to as local timing mode), instead of generating a WTOR.
Automated response to STP sync WTORs
GDPS on the controlling systems, using the BCP Internal Interface, provides automation to reply to WTOR IEA394A when the controlling systems are running in local timing mode. See “Improved controlling system availability: Enhanced timer support” on page 205. A server in an STP network might have recovered from an unsynchronized to a synchronized timing state without client intervention. By automating the response to the WTORs, potential time outs of subsystems and applications in the client’s enterprise might be averted, thus potentially preventing a production outage.
If WTOR IEA394A is posted for production systems, GDPS uses the BCP Internal Interface to automatically reply RETRY to the WTOR. If z/OS determines that the CPC is in a synchronized state, either because STP recovered or the CTN was reconfigured, it will no longer spin and continue processing. If the CPC is still in an unsynchronized state when GDPS automation responded with RETRY to the WTOR, however, the WTOR will be reposted.
The automated reply for any given system is retried for 60 minutes. After 60 minutes, you will need to manually respond to the WTOR.
7.2.2 Single-site workload configuration
A GDPS/MTMM single-site workload environment typically consists of a multisite sysplex, with all production systems running in a single site, normally Site1, and the GDPS controlling system in Site2. The controlling system (or systems, because you may have two in some configurations) will normally run in the site containing the secondary disk volumes.
The multisite sysplex can be a base sysplex or a Parallel Sysplex; a coupling facility is not strictly required. The multisite sysplex must be configured with redundant hardware (for example, a coupling facility and a Sysplex Timer in each site), and the cross-site connections must also be redundant. Instead of using Sysplex Timers to synchronize the servers, you can also use Server Time Protocol (STP) to synchronize the servers.
Figure 7-2 shows a typical GDPS/MTMM single-site workload configuration. LPARs P1 and P2 are in the production sysplex, as are the coupling facilities CF1, CF2, and CF. The primary (H1) disks are in Site1, with a set of secondaries (H2) also in Site1 and another set of secondaries (H3) in Site2. All the production systems are running in Site1, with only the GDPS controlling system (K1) running in Site2. You will notice that system K1’s disks (those marked K1) are also in Site2.
The GDPS/MTMM code itself runs under NetView and System Automation, and runs in every system in the GDPS sysplex.
Figure 7-2 GDPS/MTMM single site workload configuration
7.2.3 Multisite workload configuration
A multisite workload configuration, shown in Figure 7-3, differs from a single-site workload in that production systems are running in both sites. Although, running a multisite workload as a base sysplex is possible, seeing this configuration as a base sysplex (that is, without coupling facilities) is unusual. This is because a multisite workload is usually a result of higher availability requirements, and Parallel Sysplex and data sharing are core components of such an environment.
Because in this example we have production systems in both sites, we need to provide the capability to recover from a failure in either site. So, in this case, there is also a GDPS controlling system with its own local (not mirrored) disk running in Site1, namely System K2. Therefore, if there is a disaster that disables Site2, there will still be a GDPS controlling system available to decide how to react to that failure and what recovery actions are to be taken.
Figure 7-3 GDPS/MTMM multisite workload configuration
7.2.4 Business Recovery Services (BRS) configuration
A third configuration is known as the BRS configuration, and is illustrated in Figure 7-4 on page 209. In this configuration, all the systems in the GDPS configuration, including the controlling system, are in a sysplex in the same site, namely Site1. The sysplex does not span the two sites. The second site, Site2, might be a client site or might be owned by a third-party recovery services provider; thus the name BRS.
Site2 will contain the secondary disks and the alternate couple data sets (CDS), and might also contain processors that will be available in case of a disaster, but are not part of the configuration. This configuration can also be used when the distance between the two sites exceeds the distance supported for a multisite sysplex, but is within the maximum distance supported by FICON and Metro Mirror.
Even though there is no need for a multisite sysplex with this configuration, you must have channel connectivity from the GDPS systems to the secondary disk subsystems. Also, as explained in the next paragraph, the controlling system in Site1 will need channel connectivity to its disk devices in Site2. Therefore, FICON link connectivity from Site1 to Site2 will be required. See 2.9.7, “Connectivity options” on page 47, and IBM z Systems Connectivity Handbook, SG24-5444, for options available to extend the distance of FICON links between sites.
In the BRS configuration one of the two controlling systems must have its disk devices in Site2. This permits that system to be restarted manually in Site2 after a disaster is declared. After it restarts in Site2, the system runs a GDPS script to recover the secondary disk subsystems, reconfigure the recovery site, and restart the production systems from the disk subsystems in Site2.
If you have only a single controlling system and you have a total cross-site fiber connectivity failure, the controlling system running on Site2 disks might not be able to complete the Freeze operation because it will lose access to its disk in Site2. Having a second controlling system running on Site1 local disks in will guarantee that the freeze operation completes successfully in the event the controlling system running on Site2 disks is down or is unable to function because of a cross-site fiber loss. GDPS will attempt to maintain the current Master system in the controlling system by using the secondary disks.
Figure 7-4 GDPS/MTMM BRS configuration
7.2.5 Combining GDPS/MTMM with GDPS/XRC
GDPS/MTMM supports the existence of an additional XRC leg to be configured using the PPRC primary disk. In such a configuration, XRC Incremental Resynchronization (IR) is not supported.
If a HyperSwap or a recovery is performed on one of the PPRC legs, you can establish XRC from the new primary, provided that you have connectivity from the new primary devices to the XRC recovery region. However, a full initial copy will be required.
7.2.6 Combining GDPS/MTMM with GDPS/GM in a 4-site configuration
GDPS/MTMM (managing a single synchronous replication leg) can be combined with GDPS/GM in 3-site and 4-site configurations. In such configurations, GDPS/MTMM (when combined with Parallel Sysplex use and HyperSwap) in one region provides continuous availability across a metropolitan area or within the same local site, and GDPS/GM provides disaster recovery capability using a remote site in a different region.
The 4-site environment is configured in a symmetric manner so that there is a GDPS/MTMM-managed replication leg available in both regions to provide continuous availability (CA) within the region, with GDPS/GM to provide cross-region DR, no matter in which region production is running at any time.
This combination is referred to as GDPS/MGM Multi-Target.
7.2.7 Other considerations
The availability of the dedicated GDPS controlling system (or systems) in all scenarios is a fundamental requirement in GDPS. Merging the function of the controlling system with any other system that accesses or uses the primary volumes is not possible.
Equally important is that certain functions (stopping and restarting systems and changing the couple data set configuration) are done through the scripts and panel interface provided by GDPS. Because events such as systems going down or changes to the couple data set configuration are indicators of a potential disaster, such changes must be initiated using GDPS functions so that GDPS understands that these are planned events.
7.3 Multiplatform Resiliency for System z (also known as xDR)
To reduce IT costs and complexity, many enterprises are consolidating open servers into Linux on System z servers. Linux on System z systems can be implemented either as guests running under z/VM or native Linux on System z systems. Several examples exist of an application server running on Linux on System z and a database server running on z/OS. Two examples are as follows:
•WebSphere Application Server running on Linux and CICS, DB2 running under z/OS
•SAP application servers running on Linux and database servers running on z/OS
With a multitiered architecture, there is a need to provide a coordinated near-continuous availability and disaster recovery solution for both z/OS and Linux on System z. The GDPS/MTMM function that provides this capability is called Multiplatform Resiliency for System z, and it can be implemented if the disks being used by z/VM and Linux are CKD disks. For more details about this function, see 10.2, “GDPS/PPRC Multiplatform Resiliency for z Systems” on page 299.
Note that only Linux on System z systems implemented as guests running under z/VM are supported in GDPS/MTMM environments; native Linux on System z systems are not supported.
7.4 Managing the GDPS environment
We have seen how GDPS/MTMM can protect just about any type of data that can reside in a disk subsystem. It can also provide data consistency across multiple platforms. However, as discussed in Chapter 1, “Introduction to business resilience and the role of GDPS” on page 1, the overwhelming majority of System z outages are not disasters. Most are planned outages, with a small percentage of unplanned ones.
In this section, we describe the other aspect of GDPS/MTMM, that is, its ability to monitor and manage the resources in its environment. GDPS provides two mechanisms to help you manage the GDPS sysplex and resources within that sysplex. One mechanism is the NetView interface and the other is support for scripts. We review both of these mechanisms here.
7.4.1 NetView interface
The user interface for GDPS/MTMM is called the NetView 3270 panel interface. An example of the main GDPS/MTMM panel is shown in Figure 7-5.
Figure 7-5 GDPS/MTMM Main Panel (VPCPPNLN)
This panel has a summary of configuration status at the top, and a menu of selectable choices. As an example, to view the disk mirroring (Dasd Remote Copy) panels enter 1 at the Selection prompt, and then press Enter.
Monitoring function: Status Display Facility
GDPS also provides many monitors to check the status of disks, sysplex resources, and other GDPS-managed resources. Any time there is a configuration change, or something in GDPS that requires manual intervention, GDPS will raise an alert. GDPS uses the Status Display Facility (SDF) provided by System Automation as the primary status feedback mechanism for GDPS. It is the only dynamically updated status display available for GDPS.
GDPS provides a dynamically updated color-coded SDF panel, as shown in Figure 7-6. If something changes in the environment that requires attention, the color of the associated field on the panel will change. At all times, the operators need to have an SDF panel within view so they will immediately become aware of anything requiring intervention or action.
Figure 7-6 GDPS SDF panel
The GDPS SDF panel is divided in two parts: the top part contains status indicators, and the lower part is for trace entries. The status indicators are color-coded, with green meaning that the status is good. Minor problems are indicated by the color pink. And serious problems are shown in red. The goal is to have all status indicators green.
Remote copy panels
The z/OS Advanced Copy Services capabilities are powerful, but the native command-line interface (CLI), z/OS TSO, and ICKDSF interfaces are not as user-friendly as the GDPS DASD remote copy panels are. To more easily check and manage the remote copy environment, you can use the DASD remote copy panels provided by GDPS.
For GDPS to manage the remote copy environment, you must first define the configuration (primary and secondary LSSs, primary and secondary devices, and PPRC links) to GDPS in a file called the GEOPARM file.
After the configuration is known to GDPS, you can use the panels to check that the current configuration matches the one you want. You can start, stop, suspend, and resynchronize mirroring. These actions can be done at the device or LSS level, or both, as appropriate for a selected replication leg.
Figure 7-7 shows the Replication Leg Status and Policies panel.
Figure 7-7 DASD Remote Copy Status panel (VPCPQSTM)
The Replication Leg Status and Policies panel, is organized into three sections:
•The top section displays information related to the entire configuration, including the overall mirroring status and HyperSwap status.
•The middle section displays the replication legs, along with information related to each replication leg, including the current mirroring status, HyperSwap status, and policy information.
•The bottom section contains a list of actions that can be accessed by entering the selection number associated with each action.
View SSID Pairs panel
Entering a V (view) line command for a replication leg (on the Replication Leg Status and Policies panel, shown in Figure 7-7 on page 213) presents the panel shown in Figure 7-8.
Figure 7-8 View Storage Subsystems Status panel (VPCPQSTE)
This panel contains a lot of information and is also the place where many disk-related actions can be initiated for the selected leg. It is entirely possible that everything is working to plan on one replication leg at the same time that another replication leg is experiencing problems.
If you are familiar with using the TSO or ICKDSF interfaces, you might appreciate the ease of use of the DASD remote copy panels.
Remember that these panels provided by GDPS are not intended to be a remote copy monitoring tool. Because of the overhead involved in gathering the information for every device to populate the NetView panels, GDPS gathers this data only on a timed basis, or on demand following an operator instruction. The normal interface for finding out about remote copy status or problems is the Status Display Facility (SDF).
Standard Actions
GDPS provides facilities to help manage many common system-related planned actions. There are two reasons to use the GDPS facilities to perform these actions known as Standard Actions:
•They are well tested and based on IBM preferred procedures.
•Using the GDPS interface lets GDPS know that the changes that it is seeing (for example, a system being partitioned out of the sysplex) are planned changes, and therefore GDPS is not to react to these events.
Standard Actions are single-step actions, or are intended to impact only one resource. Examples are starting a system IPL, maintaining the various IPL address and load parameters that can be used to IPL a system, selecting the IPL address and load parameters to be used the next time a system IPL is performed, or activating/deactivating an LPAR.
If you want to stop a system, change its IPL address, then perform an IPL, you initiate three separate Standard Actions, one after the other. GDPS scripting is a facility that is suited to multi-step, multi-system actions.
The GDPS/MTMM Standard Actions panel is shown in Figure 7-9. It displays all the systems being managed by GDPS/MTMM, and for each one it shows the current status and various IPL information. To perform actions on each system, you simply use a line command letter (L to load, X to reset and so on) next to the selected system.
Figure 7-9 GDPS/MTMM Standard Actions panel (VPCPSTD1)
GDPS supports taking a stand-alone dump using the GDPS Standard Actions panel. Clients using GDPS facilities to perform HMC actions no longer need to use the HMC for taking stand-alone dumps.
Sysplex resource management
There are certain resources that are vital to the health and availability of the sysplex. In a multisite sysplex, it can be quite complex trying to manage these resources to provide the required availability while ensuring that any changes do not introduce a single point of failure.
The GDPS/MTMM Sysplex Resource Management panel, shown in Figure 7-10 on page 216, provides you with the ability to manage the resources, with knowledge about where the resources exist. For example, normally you have Primary Couple Datasets (CDS) in Site1, and your alternates in Site2. However, if you will be shutting down Site1, you still want to have a Primary and Secondary set of CDS, but both must be in Site2. The GDPS Sysplex Resource Management panels provide this capability, without you having to know specifically where each CDS is located.
GDPS provides facilities to manage coupling facilities (CFs) in your sysplex. These facilities allow for isolating all of your structures in the CF or CFs in a single site and returning to your normal configuration with structures spread across (and possibly duplexed across) the CFs in the two sites.
Isolating structures into CFs in one site, or returning to normal use with structures spread across CFs in both sites, can be accomplished through the GDPS Sysplex Resource Management panel interface or GDPS scripts. This provides an automated means for managing CFs for planned and unplanned site or disk subsystem outages.
The maintenance mode switch allows you to start or stop maintenance mode on a single CF (or multiple CFs, if all selected CFs are in the same site). DRAIN, ENABLE, and POPULATE function is still available for single CFs.
Figure 7-10 Sysplex Resource Management main menu (VPCPSPM1)
7.4.2 GDPS scripts
At this point we have shown how GDPS panels provide powerful functions to help you manage GDPS resources. However, using GDPS panels is only one way of accessing this capability. Especially when you need to initiate what might be a complex, compound, multistep procedure involving multiple GDPS resources, it is much simpler to use a script which, in effect, is a workflow.
Nearly all of the main functions that can be initiated through the GDPS panels are also available using GDPS scripts. Scripts also provide additional capabilities that are not available using the panels.
A “script” is simply a procedure recognized by GDPS that pulls together one or more GDPS functions. Scripts can be initiated manually for a planned activity through the GDPS panels (using the Planned Actions interface), automatically by GDPS in response to an event (HyperSwap), or through a batch interface. GDPS performs the first statement in the list, checks the result, and only if it is successful, proceeds to the next statement. If you perform the same steps manually, you would have to check results, which can be time-consuming, and initiate the next action. With scripts, the process is automated.
Scripts can easily be customized to automate the handling of various situations, both to handle planned changes and unplanned situations. This is an extremely important aspect of GDPS. Scripts are powerful because they can access the full capability of GDPS. The ability to invoke all the GDPS functions through a script provides the following benefits:
•Speed
The script will execute the requested actions and check results at machine speeds. Unlike a human, it does not need to search for the latest procedures or the commands manual.
•Consistency
If you were to look into most computer rooms immediately following a system outage, what would you see? Mayhem, with operators frantically scrambling for the latest system programmer instructions. All the phones ringing. Every manager within reach asking when the service will be restored. And every systems programmer with access vying for control of the keyboards. All this results in errors because humans naturally make mistakes when under pressure. But with automation, your well-tested procedures will execute in exactly the same way, time after time, regardless of how much you shout at them.
•Thoroughly tested procedures
Because they behave in a consistent manner, you can test your procedures over and over until you are sure they do everything that you want, in exactly the manner that you want. Also, because you need to code everything and cannot assume a level of knowledge (as you might with instructions intended for a human), you are forced to thoroughly think out every aspect of the action the script is intended to undertake. And because of the repeatability and ease of use of the scripts, they lend themselves more easily to frequent testing than manual procedures.
Planned Actions
As mentioned earlier, GDPS scripts are simply procedures that pull together into a list one or more GDPS functions. For the scripted procedures that you might use for a planned change, these scripts can be initiated from the panels called Planned Actions (option 6 on the main GDPS panel as shown in Figure 7-5 on page 211).
As one example, you can have a short script that stops a system and then re-IPLs it in an alternate LPAR location, as shown in Example 7-1. The sample also handles deactivating the original LPAR after the system is stopped and activating the alternate LPAR before the system is IPLed in this location.
Example 7-1 Sample script to re-IPL a system
COMM=’Example script to re-IPL system SYS1 on alternate ABNORMAL LPAR location’
SYSPLEX=’STOP SYS1’
SYSPLEX=’DEACTIVATE SYS1’
IPLTYPE=’SYS1 ABNORMAL’
SYSPLEX=’ACTIVATE SYS1 LPAR’
SYSPLEX=’LOAD SYS1’
Figure 7-11 GDPS/MTMM Planned Action
A more complex example of a Planned Action is shown in Figure 7-11. In this example, a single action in GDPS executing a planned script of only a few lines results in a complete planned site switch. Specifically, the following actions are done by GDPS:
•The systems in Site1, P1 and P3, are stopped (P2 and P4 remain active in this example).
•The sysplex resources (CDS and CF) are switched to use only those in Site2.
•A HyperSwap is executed to use the disk in Site2 (H3 disk). As a result of the swap GDPS automatically switches the IPL parameters (IPL address and load parameters) to reflect the new configuration.
•The IPL location for the P1 and P3 systems are changed to the backup LPAR location in Site2.
•The backup LPAR locations for P1 and P3 systems are activated.
•P1 and P3 are IPLed in Site2 using the disk in Site2.
Using GDPS removes the reliance on out-of-date documentation, provides a single repository for information about IPL addresses and load parameters, and ensures that the process is done the same way every time with no vital steps accidentally overlooked.
STP CTN role reassignments: Planned operations
GDPS provides a script statement that allows you to reconfigure an STP-only CTN by reassigning the STP-only CTN server roles. In an STP CTN servers (CPCs) are assigned special roles to identify which CPC is preferred to be the clock source (Preferred Time Server, or PTS), which CPC is able to take over as the clock source for planned and unplanned events (Backup Time Server, or BTS), which CPC is the active clock source (Current Time Server, or CTS), and which CPC assists in STP recovery (Arbiter).
It is strongly recommended that the server roles be reassigned before performing planned disruptive actions on any of these special role servers. Examples of planned disruptive actions are power-on reset (POR) and Activate/Deactivate. The script statement can be integrated as part of your existing control scripts to perform these planned disruptive actions.
For example, if you are planning to deactivate the CPC that is the PTS/CTS, you can now execute a script to perform the following tasks:
•Reassign the PTS/CTS role to a different CPC in the CTN
•Optionally also reassign the BTS and Arbiter roles if required
•Execute script statements you might already have in place today to deactivate the PTS/CTS CPC
After the disruptive action is completed you can execute a second script to restore the STP roles to their normal operational state, as listed here:
•Script statement to activate the CPC
•Reassign the STP server roles to their normal operational state
•Statements you might already have in existing scripts to perform IPLs and so on
Post swap scripts
These are scripts, also known as Takeover Scripts, that are intended to define actions that GDPS will execute after an unplanned HyperSwap. There are a number of specific unplanned HyperSwap scenarios and for each one, there is a reserved name for the associated Takeover script. In the case of an unplanned HyperSwap trigger, GDPS/MTMM will immediately and automatically execute an unplanned HyperSwap. Following the HyperSwap operation, GDPS will then execute the appropriate takeover script if it has been defined.
The post swap Takeover scripts have reserved names which helps GDPS determine the applicability of the script for the given unplanned swap situation. For example, if there is an unplanned swap from H1 to H3, GDPS will, if you have defined it, automatically schedule a script named SWAPSITE13.
As previously mentioned, these scripts provide you with the facility to automatically perform actions that you may want to take following an unplanned HyperSwap. Typical actions you may want to perform following an unplanned HyperSwap include resynchronizing mirroring for the MTIR replication leg and changing the couple data set configuration. For HyperSwap operations that swap production from one site to another, you might want to reconfigure STP to keep the CTS role on the CPC that is in the same site as the swapped-to, new primary devices.
Scripts for other unplanned events
GDPS monitors data-related events and also performs system-related monitoring. When GDPS detects that a z/OS system is no longer active, it verifies whether the policy definition indicates that Auto IPL has been enabled, that the threshold of the number of IPLs in the predefined time window has not been exceeded, and that no planned action is active. If these conditions are met, GDPS can automatically re-IPL the system in place, bring it back into the Parallel Sysplex, and restart the application workload (Figure 7-12).
Figure 7-12 Recovering a failed image
Although Auto IPL processing takes place automatically based on policy and does not require a script, you can have scripts prepared to provide recovery for similar events, such as a complete processor failure. In such a script, you would want to activate backup partitions for all the systems on that processor, activate CBU if appropriate, and IPL these systems. You could have one such script prepared in advance for every server in your configuration.
STP CTN role reassignments: Unplanned failure
If a failure condition has resulted in the PTS, BTS, or Arbiter no longer being an operational synchronized CPC in the CTN, a suggestion is that after the failure and possible STP recovery action, the STP roles be reassigned to operational CPCs in the CTN. The reassignment reduces the potential for a sysplex outage in the event a second failure or planned action affects one of the remaining special role CPCs.
The script statement capability described in “STP CTN role reassignments: Planned operations” on page 219 can be used to integrate the STP role reassignment as part of an existing script and eliminate the requirement for the operator to perform the STP reconfiguration task manually at the HMC.
STP WTOR IEA394A response: Unplanned failure
As described in “Improved controlling system availability: Enhanced timer support” on page 205, a loss of synchronization with the sysplex timing source will generate a disabled console WTOR. This suspends all processing on the LPAR until a response to the WTOR is provided. The WTOR message is IEA394A if the CPC is in STP timing mode (either in an STP Mixed CTN or STP-only CTN).
GDPS, using scripts, can reply (either ABORT or RETRY) to the IEA394A sync WTOR for STP on systems that are spinning because of a loss of synchronization with their Current Time Source. As described in “Automated response to STP sync WTORs” on page 206, autonomic function exists to reply RETRY automatically for 60 minutes on any GDPS systems that have posted this WTOR.
The script statement complements and extends this function, as described:
•It provides the means to reply to the message after the 60-minute automatic reply window expires.
•It can reply to the WTOR on systems that are not GDPS systems (foreign systems) that are defined to GDPS; the autonomic function only replies on GDPS systems.
•It provides the ability to reply ABORT on any systems you do not want to restart for a given failure scenario before reconfiguration and synchronization of STP.
Batch scripts
GDPS also provides a flexible batch interface to invoke planned action scripts. These scripts can be invoked:
•As a REXX program from a user terminal
•By using the IBM MVS MODIFY command to the NetView task
•From timers in NetView
•Triggered through the SA automation tables
This capability, along with the Query Services interface described in ????, provides a rich framework for user-customizable systems management procedures.
7.4.3 System Management actions
Most of the GDPS Standard Actions require actions to be done on the HMC. The interface between GDPS and the HMC is through the BCP Internal Interface (BCPii), and this allows GDPS to communicate directly with the hardware for automation of HMC actions such as Load, Stop (graceful shutdown), Reset, Activate LPAR, and Deactivate LPAR. GDPS can also perform ACTIVATE (power-on reset), CBU ACTIVATE/UNDO, OOCoD ACTIVATE/UNDO, and STP role reassignment actions against an HMC object that represents a CPC.
The GDPS LOAD and RESET Standard Actions (available through the Standard Actions panel or the SYSPLEX script statement) allow specification of a CLEAR or NOCLEAR operand. This provides operational flexibility to accommodate client procedures, thus eliminating the requirement to use the HMC to perform specific LOAD and RESET actions.
Furthermore, when you LOAD a system using GDPS (panels or scripts), GDPS can listen for operator prompts from the system being IPLed and reply to such prompts. GDPS provides support for optionally replying to such IPL-time prompts automatically, removing reliance on operator skills and eliminating operator error for selected messages that require replies.
SYSRES Management
Today many clients maintain multiple alternate z/OS SYSRES devices (also known as IPLSETs) as part of their maintenance methodology. GDPS provides special support to allow clients to identify IPLSETs. This removes the requirement for clients to manage and maintain their own procedures when IPLing a system on a different alternate SYSRES device.
GDPS can automatically update the IPL pointers after any disk switch or disk recovery action that changes the GDPS primary disk location indicator for PPRC disks. This removes the requirement for clients to perform additional script actions to switch IPL pointers after disk switches, and greatly simplifies operations for managing alternate SYSRES “sets.”
7.5 GDPS/MTMM monitoring and alerting
The GDPS SDF panel, discussed in “Monitoring function: Status Display Facility” on page 211, is where GDPS dynamically displays color-coded alerts.
Alerts can be posted as a result of an unsolicited error situation that GDPS listens for. For example, if one of the multiple PPRC links that provide the path over which PPRC operations take place is broken, there is an unsolicited error message issued. GDPS listens for this condition and will raise an alert on the SDF panel, notifying the operator of the fact that a PPRC link is not operational. Clients run with multiple PPRC links and if one is broken, PPRC continues over any remaining links. However, it is important for operations to be aware that a link is broken and fix this situation because a reduced number of links results in reduced PPRC bandwidth and reduced redundancy. If this problem is not fixed in a timely manner and more links fail, it can result in production impact because of insufficient mirroring bandwidth or total loss of PPRC connectivity (which results in a freeze).
Alerts can also be posted as a result of GDPS periodically monitoring key resources and indicators that relate to the GDPS/MTMM environment. If any of these monitoring items are found to be in a state deemed to be not normal by GDPS, an alert is posted on SDF.
Various GDPS monitoring functions are executed on the GDPS controlling systems and on the production systems. This is because, from a software perspective, it is possible that different production systems have different views of some of the resources in the environment, and although status can be normal in one production system, it can be not normal in another. All GDPS alerts generated on one system in the GDPS sysplex are propagated to all other systems in the GDPS. This propagation of alerts provides for a single focal point of control. It is sufficient for the operator to monitor SDF on the master controlling system to be aware of all alerts generated in the entire GDPS complex.
When an alert is posted, the operator will have to investigate (or escalate, as appropriate) and corrective action will need to be taken for the reported problem as soon as possible. After the problem is corrected, this is detected during the next monitoring cycle and the alert is cleared by GDPS automatically.
GDPS/MTMM monitoring and alerting capability is intended to ensure that operations are notified of and can take corrective action for any problems in their environment that can affect the ability of GDPS/MTMM to do recovery operations. This will maximize the chance of achieving your availability and RPO/RTO commitments.
7.5.1 GDPS/MTMM health checks
In addition to the GDPS/MTMM monitoring described, GDPS provides health checks. These health checks are provided as a plug-in to the z/OS Health Checker infrastructure to check that certain settings related to GDPS adhere to preferred practices.
The z/OS Health Checker infrastructure is intended to check a variety of settings to determine whether these settings adhere to z/OS optimum values. For settings found to be not in line with preferred practices, exceptions are raised in the Spool Display and Search Facility (SDSF). If these settings do not adhere to recommendations, this can hamper the ability of GDPS to perform critical functions in a timely manner.
Often, if there are changes in the client environment, this might necessitate adjustment of various parameter settings associated with z/OS, GDPS, and other products. It is possible that you can miss making these adjustments, which can affect GDPS. The GDPS health checks are intended to detect such situations and avoid incidents where GDPS is unable to perform its job because of a setting that is perhaps less than ideal.
For example, GDPS/MTMM provides facilities for management of the couple data sets (CDS) for the GDPS sysplex. One of the health checks provided by GDPS/MTMM checks that the couple data sets are allocated and defined to GDPS in line with the GDPS preferred practices recommendations.
Similar to z/OS and other products that provide health checks, GDPS health checks are optional. Several optimum values that are checked and the frequency of the checks can be customized to cater to unique client environments and requirements.
There are a few z/OS preferred practices that conflict with GDPS preferred practices. The related z/OS and GDPS health checks result in conflicting exceptions being raised. For such health check items, to avoid conflicting exceptions, z/OS provides the ability to define a coexistence policy where you can indicate which practice is to take precedence; GDPS or z/OS. GDPS provides sample coexistence policy definitions for the GDPS checks that are known to be conflicting with z/OS.
GDPS also provides a convenient interface for managing the health checks using the GDPS panels. You can use it to perform actions such as activate/deactivate or run any selected health check, view the customer overrides in effect for any optimum values, and so on. Figure 7-13 shows a sample of the GDPS Health Checks Information Management panel. In this example you see that all the health checks are enabled. The status of the last run is also shown, indicating that some were successful and some resulted in raising a medium exception. The exceptions can also be viewed using other options on the panel.
Figure 7-13 GDPS/MTMM Health Checks Information Management panel (VPC8PHC0)
7.6 Other facilities related to GDPS
Miscellaneous facilities that GDPS/MTMM provides can assist in various ways, such as reducing the window during which disaster recovery capability is not available.
7.6.1 HyperSwap and TDMF coexistence
To minimize disruption to production workloads and service levels, many enterprises use IBM’s Transparent Data Migration Facility (TDMF) for storage subsystem migrations and other disk relocation activities. The migration process is transparent to the application, and the data is continuously available for read and write activities throughout the migration process.
However, the HyperSwap function is mutually exclusive with software that moves volumes around by switching UCB pointers. The currently supported versions of TDMF and GDPS allow operational coexistence. With this support, TDMF automatically temporarily disables HyperSwap as part of the disk migration process only during the brief time when it switches UCB pointers. Manual operator interaction is not required. Without this support, through operator intervention, HyperSwap is disabled for the entire disk migration, including the lengthy data copy phase.
7.6.2 Reduced impact initial copy and resynchronization
Performing PPRC copy of a large amount of data across a large number of devices while the same devices are used in production by application workloads can potentially affect production I/O service times if such copy operations are performed synchronously. Your disk subsystems and PPRC link capacity are typically sized for steady state update activity, but not for bulk, synchronous replication. Initial copy of disks and resynchronization of disks are examples of bulk copy operations that can affect production if performed synchronously.
There is no need to perform initial copy or resynchronizations using synchronous copy, because the secondary disks cannot be made consistent until all disks in the configuration have reached duplex state.
GDPS supports initial copy and resynchronization using asynchronous PPRC-XD (also known as Global Copy). When GDPS initiates copy operations in asynchronous copy mode, GDPS monitors progress of the copy operation and when the volumes are near full duplex state, GDPS converts the replication from the asynchronous copy mode to synchronous PPRC. Initial copy or resynchronization using PPRC-XD eliminates the performance impact of synchronous mirroring on production workloads.
Without asynchronous copy it might be necessary to defer these operations or reduce the number of volumes being copied at any given time. This would delay the mirror from reaching a duplex state, thus impacting a client’s ability to recovery. Use of the XD-mode asynchronous copy allows clients to establish or resynchronize mirroring during periods of high production workload, and can potentially reduce the time during which the configuration is exposed.
7.6.3 Concurrent Copy cleanup
The DFSMS Concurrent Copy (CC) function uses a “sidefile” that is kept in the disk subsystem cache to maintain a copy of changed tracks that have not yet been copied. For a PPRCed disk, this sidefile is not mirrored to the secondary subsystem. If you perform a HyperSwap while a Concurrent Copy operation is in progress, this will result in the job performing the copy failing after the swap. GDPS will not allow a planned swap when a Concurrent Copy session exists against your primary PPRC devices. However, unplanned swaps will still be allowed.
If you plan to use HyperSwap for primary disk subsystem failures (unplanned HyperSwap), try to eliminate any use of Concurrent Copy because you cannot plan when a failure will occur. If you choose to run Concurrent Copy operations while enabled for unplanned HyperSwaps, and a swap occurs when a Concurrent Copy operation is in progress, the job performing the Concurrent Copy operation is expected to fail.
Checking for CC is performed by GDPS immediately before performing a planned HyperSwap. SDF trace entries are generated if one or more CC sessions exist, and the swap command will end with no PPRC device pairs being swapped. You must identify and terminate any CC and XRC sessions against the PPRC primary devices before the swap.
When attempting to resynchronize your disks, checking is performed to ensure that the secondary devices do not retain CC status from the time when they were primary devices. These are not supported as PPRC secondary devices. Therefore, GDPS will not attempt to establish a duplex pair with secondary devices if it detects a CC session.
GDPS provides a function to discover and terminate Concurrent Copy sessions that would otherwise cause errors during a resync operation. The function is controlled by a keyword that provides options to disable, to conditionally enable, or to unconditionally enable the cleanup of Concurrent Copy sessions on the target disks. This capability eliminates the manual task of identifying and cleaning up orphaned Concurrent Copy sessions before resynchronizing a suspended PPRC mirror.
7.6.4 Easy Tier Heat Map Transfer
IBM DS8000 Easy Tier optimizes data placement (placement of logical volumes) across the various physical tiers of storage within a disk subsystem to optimize application performance. The placement decisions are based on learning the data access patterns, and can be changed dynamically and transparently using this data.
PPRC mirrors the data from the primary to the secondary disk subsystem. However, the Easy Tier learning information is not included in PPRC scope. The secondary disk subsystems are optimized according to the workload on these subsystems, which is different than the activity on the primary (there is only write workload on the secondary whereas there is read/write activity on the primary). As a result of this difference, during a disk switch or disk recovery, the secondary disks that you switch to are likely to display different performance characteristics compared to the former primary.
Easy Tier Heat Map Transfer is the DS8000 capability to transfer the Easy Tier learning from a PPRC primary to the secondary disk subsystems so that the secondary disk subsystems can also be optimized, based on this learning, and will have similar performance characteristics if it is promoted to become the primary.
GDPS integrates support for Heat Map Transfer. In a Multi-Target PPRC environment, Heat Map Transfer is established for both secondary targets. The appropriate Heat Map Transfer actions (such as start/stop of the processing and reversing transfer direction) are incorporated into the GDPS managed processes. For example, if PPRC is temporarily suspended on one leg by GDPS for a planned or unplanned secondary disk outage, Heat Map Transfer is also suspended on that leg, or if PPRC direction is reversed as a result of a HyperSwap, Heat Map Transfer direction is also reversed.
7.7 GDPS/MTMM flexible testing and resync protection
Configuring point-in-time copy (FlashCopy) capacity in your MTMM environment provides two significant benefits:
•You can conduct regular DR drills or other tests using a copy of production data while production continues to run.
•You can save a consistent, “golden” copy of the PPRC secondary data, which can be used if the primary disk or site is lost during a PPRC resynchronization operation.
FlashCopy and the various options related to FlashCopy are discussed in 2.6, “FlashCopy” on page 38. GDPS/MTMM supports taking a FlashCopy of the current primary or either of the current secondary disks sets. The COPY, NOCOPY, NOCOPY2COPY, and INCREMENTAL options are supported. CONSISTENT FlashCopy is supported in conjunction with COPY, NOCOPY, and INCREMENTAL FlashCopy.
FlashCopy can also be used, for example, to back up data without the need for extended outages to production systems; to provide data for data mining applications; for batch reporting, and so on.
7.7.1 Use of space-efficient FlashCopy volumes
As discussed in “Space-efficient FlashCopy (FlashCopy SE)” on page 40, by using space-efficient (SE) volumes, you might be able to lower the amount of physical storage needed and thereby reduce the cost associated with providing a tertiary copy of the data. GDPS provides support allowing space-efficient FlashCopy volumes to be used as FlashCopy target disk volumes. Whether a target device is space-efficient or not is transparent to GDPS; if any of the FlashCopy target devices defined to GDPS are space-efficient volumes, GDPS will simply use them. All GDPS FlashCopy operations with the NOCOPY option, whether through GDPS scripts, panels, or FlashCopies automatically taken by GDPS, can use space-efficient targets.
Space-efficient volumes are ideally suited for FlashCopy targets when used for resync protection. The FlashCopy is taken before the resync and can be withdrawn as soon as the resync operation is complete. As changed tracks are sent to the secondary for resync, the time zero (T0) copy of this data is moved from the secondary to the FlashCopy target device. This means that the total space requirement for the targets is equal to the number of tracks that were out of sync, which typically will be significantly less than a full set of fully provisioned disks.
Another potential use of space-efficient volumes is if you want to use the data for limited disaster recovery testing.
Understanding the characteristics of FlashCopy SE is important to determine whether this method of creating a point-in-time copy will satisfy your business requirements. For example, will it be acceptable to your business if, because of an unexpected workload condition, the repository on the disk subsystem for the space-efficient devices becomes full and your FlashCopy is invalidated so that you are unable to use it? If your business requirements dictate that the copy must always be guaranteed to be usable, space-efficient might not be the best option and you can consider using standard FlashCopy instead.
7.8 GDPS tools for GDPS/MTMM
GDPS/MTMM includes tools that provide function that is complementary to GDPS function. The tools represent the kind of function that all or many clients are likely to develop themselves to complement GDPS. Using these tools eliminates the need for you to develop similar function yourself. The tools are provided in source code format which means that if the tool does not exactly meet your requirements, you can modify the code to suit your needs.
The following tools are available with GDPS/MTMM:
•GDPS XML Conversion (GeoXML) Tool
This tool helps you to convert an existing GDPS/PPRC (or GDPS/PPRC HyperSwap Manager - GDPS/HM) GEOPARM configuration definition file for a single replication leg to GDPS/MTMM XML format GEOPARM definitions. This simplifies the task of defining the MTMM configuration for existing GDPS/PPRC (or GDPS/HM) clients that will be moving to using GDPS/MTMM.
•GDPS EasyLog Tool
This is a Microsoft Windows-based tool to help you extract and easily download the MVS Syslog and NetView log from a z/OS environment. It also helps in analyzing the Netlog after it is downloaded to a workstation.
7.9 Services component
As you have learned, GDPS affect much more than simply remote copy. It also includes system, server hardware and sysplex management, automation, testing processes, disaster recovery processes, and so on.
Most installations do not have skills in all these areas readily available. It is also extremely rare to find a team that has this range of skills across many implementations. However, the GDPS/MTMM offering includes exactly that: access to a global team of specialists in all the disciplines you need to ensure a successful GDPS/MTMM implementation.
Specifically, the Services component includes several or all of the following services:
•Planning to determine availability requirements, configuration recommendations, and implementation and testing plans
•Installation and necessary customization of NetView and System Automation
•Remote copy implementation
•GDPS/MTMM automation code installation and policy customization
•Assistance in defining Recovery Point and Recovery Time objectives
•Education and training on GDPS/MTMM setup and operations
•Onsite implementation assistance
•Project management and support throughout the engagement
The sizing of the Services component of each project is tailored for that project, based on many factors including what automation is already in place, whether remote copy is already in place, whether the two centers are already in place with a multisite sysplex, and so on. This means that the skills provided are tailored to the specific needs of each particular implementation.
7.10 GDPS/MTMM prerequisites
See the following web page for the latest GDPS/MTMM prerequisite information:
7.11 Comparison of GDPS/MTMM versus other GDPS offerings
So many features and functions are available in the various members of the GDPS family that recalling them all and remembering which offerings support them is sometimes difficult. To position the offerings, Table 7-1 lists the key features and functions and indicates which ones are delivered by the various GDPS offerings.
Table 7-1 Supported features matrix
Feature
GDPS/PPRC
GDPS/PPRC HM
GDPS/MTMM
GDPS Virtual Appliance
GDPS/XRC
GDPS/GM
Continuous availability
Yes
Yes
Yes
Yes
No
No
Disaster recovery
Yes
Yes
Yes
Yes
Yes
Yes
CA/DR protection against multiple failures
No
No
Yes
No
No
No
Continuous Availability for foreign z/OS systems
Yes with z/OS Proxy
No
No
No
No
No
Supported distance
200 km
300 km (BRS configuration)
200 km
300 km (BRS configuration)
200 km
300 km (BRS configuration)
200 km
300 km (BRS configuration)
Virtually unlimited
Virtually unlimited
Zero Suspend FlashCopy support
Yes, using CONSISTENT
Yes, using CONSISTENT for secondary only
Yes, using CONSISTENT
No
Yes, using Zero Suspend FlashCopy
Yes, using CGPause
Reduced impact initial copy/resync
Yes
Yes
Yes
Yes
Not applicable
Not applicable
Tape replication support
Yes
No
No
No
No
No
Production sysplex automation
Yes
No
Yes
Not applicable
No
No
Span of control
Both sites
Both sites
(disk only)
Both sites
Both sites
Recovery site
Disk at both sites; recovery site (CBU or LPARs)
GDPS scripting
Yes
No
Yes
Yes
Yes
Yes
Monitoring, alerting and health checks
Yes
Yes
Yes
Yes (except health checks)
Yes
Yes
Query Services
Yes
Yes
No
No
Yes
Yes
MSS support for added scalability
Yes (secondary in MSS1)
Yes (secondary in MSS1)
Yes (H2 in MSS1, H3 in MSS2)
No
No
Yes (GM FC and Primary for MGM in MSS1)
MGM 3-site and 4-site
Yes (all configurations)
Yes (3-site only and non-IR only)
Yes (all configurations)
No
Not applicable
Yes (all configurations)
MzGM
Yes
Yes
Yes (non-IR only)
No
Yes
Not applicable
Open LUN
Yes
Yes
No
No
No
Yes
z/OS equivalent function for Linux for IBM z Systems
Yes
No
Yes (Linux for IBM z Systems running as a z/VM guest only)
Yes (Linux for IBM z Systems running as a z/VM guest only)
Yes
Yes
Heterogeneous support through DCM
Yes (VCS and SA AppMan)
No
No
No
Yes (VCS only)
Yes (VCS and SA AppMan)
z/BX hardware management
Yes
No
No
No
No
No
GDPS GUI
Yes
Yes
No
Yes
No
Yes
7.12 Summary
GDPS/MTMM is a powerful offering that provides disaster recovery, continuous availability, and system/sysplex resource management capabilities. HyperSwap, available with GDPS/MTMM, provides the ability to transparently swap disks between disk locations. The power of automation allows you to test and perfect the actions to be taken, either for planned or unplanned changes, thus minimizing or eliminating the risk of human error.
This offering is one of the offerings in the GDPS family, along with GDPS/PPRC, GDPS/HM, and GDPS Virtual Appliance, that offers the potential of zero data loss, and that can achieve the shortest recovery time objective, typically less than one hour following a complete site failure.
It is also one of the only members of the GDPS family, along with GDPS/PPRC and GDPS Virtual Appliance, that is based on hardware replication and that provides the capability to manage the production LPARs. Although GDPS/XRC and GDPS/GM offer LPAR management, their scope for system management only includes the systems in the recovery site, and not the production systems running in Site1.
GDPS/MTMM is the only GDPS offering that can provide zero-data-loss disaster recovery protection, even after a primary disk failure.
In addition to the disaster recovery and planned reconfiguration capabilities, GDPS/MTMM also provides a user-friendly interface for monitoring and managing the various elements of the GDPS configuration.
..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset