What is the Impact of Cache write hit rate less than 100% on Host ?
Write hit rates less than 100% have no appreciable impact on performance in most instances.
Write hit rates only indicate if there was a free block in cache to accept the write. If there are no free blocks, then a clean block is used.
Free=No data, Clean=Data that is already destaged to disk, or was prefetched.
We had a discussion that involved this question a long time ago:
Tuning Manager Response Time
I think what you're looking for was in the final post in this discussion. To paraphrase myself:
It is platform dependent. It will always be 100% on a enterprise arrays (hence they N/A), but not on modular. In a modular array If the write pending rate is high enough, you will run out of cache slots and the writes are delayed. This is reflected as a drop in write hit. Although under normal operations all writes do hit cache - the other considers here are, what happens during a memory reconfiguration in Modular/entry enterprise or a failure in enterprise and are also write-through scenarios on both (but we don't need to go into that here).
So in modular the metric can be used to indicate cache stress due to backend disk congestion (or misconfiguration, mainly around DDO and DDSO), but this is not the case in enterprise.
We have a VSP array and there are many ldevs where write hit rate is 0% to 50% with less than 10% CWP.
What shud I conclude?????
Do I need to increase cache size?
Will inflow control come into picture if CWP is 10% and Ldev cache write hit rate is less than 100%. ????
I think you're looking at cache write hit (which is a ratio). This is really more commonly known as the re-hit ratio and it denotes if a write was already in cache. I don't think there is expected or recommended number for this metric as it depends on the locality of the write and how skewed the data set is.
What if there are no free cache slots, clean slots available and data is also not prefetched ???
I want to know
1) when will inflow control come into picture and
2) how will we know that the delays experienced by hosts is due to inflow control ?
The only thing you need to monitor is Cache Write Pending. Write Hit is of no real value. Clean Slots are used for writes all the time. It is normal.
Q1) when will inflow control come into picture
A1) For VSP, beginning at Write Pending above 40%. As it increases beyond 40% inflow penalties increase in scope and magnitude. At 65% write pending it reaches maximum impact, and performance is severely degraded.
Q2) how will we know that the delays experienced by hosts is due to inflow control ?
A2) Write Pending is above 40%. Write response is elevated.
Retrieving data ...