Why should an MTU value of at least 2,500 bytes be configured in a bridged iSCSI environment?

The MTU of a network interface determines the maximum size of the packets that can be transmitted without IP fragmentation. Ethernet uses a frame size of 1,500 bytes per frame by default. In bridged iSCSI environment, FC to IP and vice versa translation occurs at the bridge. Since FC frame size is 2148 bytes, FC frames must be split in two parts, and each of them must be sent to the iSCSI initiator. This results in two TCP segments for each Fibre Channel frame.

Multiple frames means TCP/IP stacks overhead in the CPU because the data is divided into multiple packets, checksums are calculated for each, and packets must be reassembled at the destination. Small packets also consume network bandwidth because more time is spent in inter-frame gaps, frame headers and checksums relative to data.

MTU can be adjusted to take advantage of jumbo frames support. Using jumbo frames, an iSCSI payload can be configured to completely fit a full-size Fibre Channel frame (2148 bytes), thereby eliminating the requirement to fragment the frame into two TCP packets. In order to use jumbo frames support, the entire IP infrastructure between the end points must support jumbo frames.

Jumbo frames are defined as all frames that have MTUs larger than the standard, originally specified Ethernet payload size of 1500 bytes. A standard size of jumbo frame is 9000s byte and 'mini jumbo' frame is 2500 bytes.