04-06-2016 04:23 AM
We basically want to test QOS so that we properly understand it before implementing it on our live SAN. The aim is that we get latency events for various reasons, and although as time goes on we add or tune different features to mitigate or eradicate different reasons for latency, a new flavour of latency always seems to pop up somewhere else. And for various reasons, we need to start protecting certain traffic from the impact of SAN latency.
However in our testing, it would just be nice to be able to 'see QOS working' somehow.
Thanks to Alexey's suggestion, we've found you can create a flow using Flow Vision in a QOS class, so our current tactic is to try to use this create latency in the low priority QOS group, which shows up under the latency count in portsstatshow, but then also to run io generators on three servers in each QOS group, and hopefully see SCSI timeouts on the low priority server but no interuption on the others. Getting latency messages is the easy bit, actually producing C3 discards seems much harder.
04-06-2016 01:41 PM
I think you could imitate the slow drain device (SDD) by setting its port speed below the fabric value. This will lead to SDD are not being able to return credits fast enough and to credits starvation back pressure into fabric. Finaly you might see c3 discards due to timeout.