PCI Express Packet Latency Matters

Latency is the delay between starting and completing an action. For a switch, it's the time between the first bit of a packet on an input pin and the first bit of that packet on an output pin forwarded through the switch.

Latency matters when it affects higher level functional throughput. When looking at the action of forwarding a packet through a switch, one can conclude that high throughput can be achieved in a switch independent of its switch latency. However, sending a packet through a switch is always part of a response to a higher level sequence of actions and the cumulative delay of that sequence may indeed affect the throughput of the higher level operation of which it is a part.

A key example of this is a CPU centric system. In a CPU-centric system, the processor performs a sequence of reads and writes to its memory and I/Os directly or through DMA. One downside of packet latency in this system is that some otherwise productive resource of a system must idle during the latency period, waiting for either data upon which to do work or for instructions as to what work to do.

View Entire Paper | Previous Page | White Papers Search

If you found this page useful, bookmark and share it on: