This chapter introduces the concept of technical computing, the value of cloud computing, and the types of cloud for enterprises.

This chapter includes the following sections:

This section describes Technical Computing.

This section introduces the history of high-performance computing (HPC) and how Technical Computing became mainstream.

The IT Industry has always tried to maintain a balance between demands from business to deliver services against cost considerations of hardware and software assets. On one hand, business growth depends on information technology (IT) being able to provide accurate, timely, and reliable services. On other hand, there is cost associated with running IT services. These concerns have led to the growth and development of HPC.

HPC has traditionally been the domain of powerful computers (called “supercomputers”) owned by governments and large multinationals. Existing hardware was used to process data and provide meaningful information to single systems working with multiple parallel processing units. Limitations were based on hardware and software processing capabilities. Due to the cost associated with such intensive hardware, the usage was limited to a few nations and corporate entities.

The advent of the workflow-based processing model and virtualization as well as high availability concepts of clustering and parallel processing have enabled existing hardware to provide the performance of the traditional supercomputers. New technologies such as graphics processing units (GPUs) have pushed power of the existing hardware to perform more complicated functions faster than previously possible. Virtualization and clustering have made it possible to provide a greater level of complexity and availability of IT services. Sharing of resources to reduce cost has also become possible due to virtualization. There has been a move from a traditionally static IT model based on maximum load sizing to a leaner IT model based on workflow-based resource allocation through smart clusters. With the introduction of cloud technology, the resource requirement is becoming more on-demand as compared to the traditional forecasted demand, thus optimizing cost considerations further.

These technological innovations have made it possible to push the performance limits of existing IT resources to provide high performance output. The technical power to achieve computing results can be achieved with much cheaper hardware using smart clusters and grids of shared hardware. With workflow-based resource allocation, it is possible to achieve high performance from a set of relatively inexpensive hardware working together as a cluster. Performance can be enhanced by breaking across silos of IT resources, lying dormant to provide on-demand computing power wherever required. Data intensive industries such as engineering and life sciences can now use the computing power on demand provided by the workflow-based technology. Using parallel processing by heterogeneous resources that work as one unit under smart clusters, complex unstructured data can be processed to feed usable information into the system.

With the reduction in the cost of hardware resources, the demand for HPC has spread technical computing from scientific labs to mainstream commercial applications (Figure 1-1 on page 3). Technical computing has been demanded from sectors such as aerodynamics, automobile design, engineering, financial services, and oil and gas Industries. Improvement in cooling technology and power management of these superfast computing grids have allowed users to extract more efficiency and performance from existing hardware.

Increased complexity of applications and demand for faster analysis of data has led Technical Computing to become widely available. Thus, IBM Technical Computing is focused on helping clients to transform their IT infrastructure to accelerate results. The goal of Technical Computing in mainstream industries is to meet the challenges of applications that require high performance computing, faster access to data, and intelligent workload management.

The following provides a description of the terminology used in this book.

IBM Platform Computing solutions have gone through the evolution from cluster to grid to cloud due to its abilities to manage heterogeneous complexities of distributed computing resources. Figure 1-2 shows the evolution of clusters, grids, and HPC clouds.

IBM Platform Computing provides solutions for mission-critical applications that require complex workload management across heterogeneous environment for diverse industries from life sciences to engineering and financial sectors that involve complex risk analysis. IBM Platform Computing has a 20-year history of working on highly complex solutions for some of the largest multinational companies. It has proven examples of robust management of highly complex workflow across large distributed environments that deliver results.

This section provides a brief overview of the components (hardware, software, storage) available to help deploy a technical computing cloud environment. The following sections provide a subset of the possible solutions.

IBM HPC and IBM Technical Computing provide flexibility in your choice of hardware and software:

In addition to this list, IBM Platform Computing provides support to heterogeneous cluster environments with extra IBM or third-party software (Figure 1-3):

IBM Cluster Manager tools helps use the bandwidth of the network devices to lower the latency levels. The following are some of the devices supported:

IBM Cluster Manager tools use storage devices capable of high parallel I/O to help provide efficient I/O related operations in the cloud environment. The following are some of the storage devices that are used:

Technical computing workloads have the following characteristics:

The next section provides a few technologies that support technical computing workloads.

HPC clusters frequently employ a distributed memory model to divide a computational problem into elements that can be simultaneously run in parallel on the hosts of a cluster. This often involves the requirement that the hosts share progress information and partial results by using the cluster’s interconnect fabric. This is most commonly accomplished by using a message passing mechanism. The most widely adopted standard for this type of message passing is the MPI standard, which is described on the following website:

IBM Platform MPI is a high-performance and production-quality implementation of the MPI standard. It fully complies with the MPI-2.2 standard, and provides enhancements such as low latency and high bandwidth point-to-point and collective communication routines over other implementations.

For more information about IBM Platform MPI, see the IBM Platform MPI User’s Guide, SC27-4758-00, at:

SOA is a software architecture in which the business logics are encapsulated and defined as services. These services can be used and reused by one or multiple systems that participate in the architecture. SOA implementations are generally platform-independent, which means that infrastructure considerations do not get in the way of deploying new systems or enhancing existing systems. Many financial institutions deploy a range of technologies, so the heterogeneous nature of SOA is particularly important.

IBM Platform Symphony combines a fast service-oriented application middleware component with a highly scalable grid management infrastructure. Its design delivers reliability and flexibility, while also ensuring low levels of latency and high throughput between all system components.

For more information about SOA, see:

MapReduce is a programming model for applications that process large volumes of data in parallel by dividing the work into a set of independent tasks across many systems. MapReduce programs in general transform lists of input data elements into lists of output data elements in two phases: Map and reduce.

MapReduce is widely used in the data intensive computing such as business analytics and life science. Within IBM Platform Symphony, the MapReduce framework supports data-intensive workload management using a special implementation of service-oriented application middleware to manage MapReduce workloads.

Workflow is a task that is composed by a sequence of connected steps. In HPC clusters, many workflows run in parallel to complete a job or to respond to a batch of requests. As the complexity increases, workflows become more complicated. Workflow automation is becoming increasingly important for these reasons:

Clients need an easy-to-use and cost efficient way to develop and maintain the workflows.

Visualization is a typical workload in engineering for airplane and automobile designers. The designers create large computer-aided design (CAD) environments to run their 2D/3D graphic calculations and simulations for the products. These workloads demand a large hardware environment that includes graphic workstations, storage, and software tools. In addition to the hardware, the software licenses are also expensive. Thus, the designers are looking to reduced costs, and expect to share the infrastructure between computer-aided engineering (CAE) and CAD.

Implementing a cloud infrastructure can be the ideal solution for companies who do not want to invest in a separate cluster infrastructure for technical computing workloads. It can reduce, among other things, extra hardware and software costs and avoid the extra burden of another cluster administration. Cloud also provides the benefits of request on demand and release on demand after the work is completed, which saves time for deployments and the expenses to a certain extent. For technical computing, the hardware requirements are usually large considering the workloads that it must manage. Although the physical hardware runs better in HPC environments, evolving virtualization technologies have started to provide room for HPC solutions as well. Using a computing cloud for HPC environments can help eliminate the static usage of the infrastructure. It can also help provide a way to use the hardware resources dynamically as per the computing requirements.

Cloud computing provides the flexibility to use the resources when required. In terms of a technical computing cloud environment, cloud computing not only provides the flexibility to use the resources on demand, but helps to provision the computing nodes as per the application requirement to help manage the workload. By implementing and using IBM Platform Computing Manager (PCM), dynamic provisioning of the computing nodes with the wanted operating systems is easily achieved. This dynamic provisioning solution helps to better use the hardware resources and fulfill various technical computing requirements for managing the workloads. Figure 1-4 shows the infrastructure of an HPC cloud.

Cloud computing can significantly reduce manual effort during installation, provisioning, configuration, and other tasks that were performed manually before. When done manually, these computing resource management steps can take a significant amount of time. A cloud-computing environment can dramatically help reduce the system management complexity by implementing automation, business workflows, and resource abstractions.

IBM PCMAE provides many automation features to help reduce the complexity of managing a cloud-computing environment:

These automation features reduce the time that is required to make the resources available to clients.

In a cloud computing environment, many computers, network devices, storage, and applications are running. To achieve high availability, throughput, and resource utilization, clouds have monitoring mechanisms. Monitoring measures the service and resource usage, which is key for charge back to the users. The system statistics are collected and reported to the cloud provider or user, and based on these figures, dashboards can be generated.

Monitoring provides the following benefits:

IBM SmartCloud Monitoring 7.1 is a bundle of established IBM Tivoli infrastructure management products, including IBM Tivoli Monitoring and IBM Tivoli Monitoring for Virtual Environments. The software delivers dynamic usage trending and health alerts for pooled hardware resources in the cloud infrastructure. The software includes sophisticated analytics, and capacity reporting and planning tools. You can use these tools to ensure that the cloud is handling workloads quickly and efficiently.

For more information about IBM SmartCloud Monitoring, see the following website:

There are three different cloud-computing architectures:

A private cloud is an architecture where the client encapsulates its IT capacities “as a service” over an intranet for their exclusive use. The cloud is owned by the client, and is managed and hosted by the client or a third party. The client defines the ways to access the cloud. The advantage is that the client controls the cloud so that security and privacy can be ensured. Also, the client can customize the cloud infrastructure based on its business needs. A private cloud can be cost effective for a company that owns many computing resources.

A public cloud provides standardized services for public use over the Internet. Usually it is built on standard and open technologies, providing web page, API or SDK for the consumers to use the services. Benefits include standardization, capital preservation, flexibility, and improved time to deploy.

Clients can integrate a private cloud and a public cloud to deliver computing services, which is called hybrid cloud computing. Figure 1-5 highlights the differences and relationships of these three types of clouds.

IBM HPC clouds can help enable transformation of both your IT infrastructure and business. Based on an HPC cloud’s potential impact, clients are actively evolving their infrastructure toward private clouds, and beginning to consider public and hybrid clouds. Clients are transforming their existing infrastructure to HPC clouds to enhance the responsiveness, flexibility, and cost effectiveness of their environment. This transformation helps clients enable an integrated approach to improve computing resource capacity and to preserve capital. Eventually the client will access extra cloud capacity by using the cloud models described in Figure 1-5.

In a public cloud environment, HPC must overcome a number of significant challenges as shown in Table 1-1.

When using private clouds, HPC might not suffer from the public cloud barriers, but there are other common issues as shown in Table 1-2.

Figure 1-6 shows the IBM HPC cloud reference model.

The HPC private cloud has three hosting models: Private cloud, managed private cloud, and hosted private cloud. Table 1-3 describes the characteristics of these models.