Packet Addressing Basics

A packet is used to send a request or a response from one CA to another. One packet's data payload field can contain a maximum of 4096 bytes of data. An adapter sending a message that is larger than a single packet's data payload field must therefore segment the message transfer into a series of two or more packets. Each packet contains a data payload field, one or two packet routing headers, CRCs, etc. Each packet contains the following address information in its routing headers:

• Local Route Header (LRH) is present in every packet (see Figure 1-4 on page 15 and Figure 1-5 on page 15). Among other elements, the LRH contains:

  - Destination port Local ID (DLID). This 16-bit field is used by switches to guide the packet through the subnet towards the destination port in this subnet.
  
  

  - Source port Local ID (SLID). This is the 16-bit address of the source port that originates a request packet in a subnet.
  
  

  - Depending on the version of the protocol used to perform the transfer, the destination CA may respond to a request packet by returning a response packet. The SLID and DLID fields obtained from the request packet are reversed in the response packet's LRH.

  Figure 1-4. Packet Format When the Source and Destination Ports Are In the Same Subnet

  

  Figure 1-5. Packet Format When the Source and Destination Ports Are In Different Subnets

  

• Global Router Header (GRH). See Figure 1-5 on page 15. The GRH is only present if the source and destination CAs are not in the same subnet. Among other elements, the GRH contains:

  - Destination port Global ID (DGID). This field is used by routers to guide the packet through the fabric towards the subnet within which the destination port resides. It is a 128-bit value. The upper 64-bits identifies the subnet within which the destination CA port resides. The lower 64-bits is referred to as the Globally Unique ID (GUID) and uniquely identifies the destination port.
  
  

  - Source port Global ID (SGID). This is the 128-bit address of the source port that originates a request packet. The upper 64-bits identifies the subnet within which the source CA port resides. The lower 64-bits is the source port's GUID that uniquely identifies the source port.
  
  

  - Depending on the version of the protocol used to perform the transfer, the destination CA may respond to a request packet by returning a response packet. The SGID and DGID fields obtained from the request packet are reversed in the response packet's GRH.