configuration short-write

Hi Joachim,

I can definitely shed some light.

So in the past, pre-short-write, Metro-FC traffic basically needed two round-trips in order for one message to complete between VPLEX clusters.  Short-write, introduced in GeoSynchrony 5.2 SP1 (VPlex-5.2.1.00.00.04), cuts that in half and only requires essentially one round-trip per message.  Think of VPLEX Metro's short-write feature as identical in functionality to VMAX SRDF's SiRT feature (Single round-trip) or any FCIP vendor's fast-write feature.

If you have a Metro-FC environment that's not FCIP w/ fast writer acceleration, then you most definitely want to go to at least 5.2 SP1 for improved distributed-device write latency.  It has been proven to show significant improvements in write latency on write I/O's.  Bandwidth is unchanged although might end up showing a net benefit because of the faster response times (less waiting time between successive I/O's.)

Obviously for clusters that are already pretty close together (low round-trip-time) you may not notice a difference, however at larger distances (say greater than 2 msec) you will most definitely notice the difference.

I want to stress that for a FCIP environment, the vendor's FCIP fast-write acceleration feature should be disabled.  There has been inter-cluster stability issues observed when the two are enabled at the same time.  It's one or the other (VPLEX's or the switch's), not both.  VPLEX's short write feature can be disabled or viewed via the CLI ("configuration short-write", I believe introduced in GeoSynchrony 5.3.) 

(And in the past, FCIP accelration worked on Brocade switches accelerated VPLEX Metro traffic, but didn't appear to work as expected on Cisco switches.)

Gary