Low Latency Switching is More than Port to Port – It is Now about Switch to Switch Uniformity across the Network

For people who like to get in front of an emerging trend, I have one for you.  I think I am first person to write it or state it publically, but I am not the first to think about it.  When port to port switch latency was >1 µs, there was money to be made by building a better and faster switch.  Now that we are well below this level, racing to the closet position near zero is a Don Quixote quest.

I think the emerging trend is uniformity or predicable latency across the network.  I would describe it as offering nearly the same latency from a switch in rack 1 to a switch in rack 2 to a switch in rack 11 or rack 26.  View the network as system and be able to guarantee uniform latency from switch to switch for a cluster of 528x10G servers.  The trend is towards big data and higher work loads and this will drive the demand for system or cluster level latency uniformity.

/wrk

* It is all about the network stupid, because it is all about compute. *

** Comments are always welcome in the comments section or in private. **