Read and write cache settings for FATA

[Jeremy C]

EVA 8100; xcs 6.220; Command View EVA 9.3

Despite having 78 disks in FATA storage group, it doesn't take much IO (particularly writes) to overwhelm the FATA disks and get the queue depth rising. Ultimately this affects overall EVA latency since the IO requests are all bunched together for both FATA and FC disk groups.

Would setting the read cache to "off" and/or setting the write cache to "write-through" help or hinder with this?

It seems that the best solution would be if HP had a firmware update to separate the IO requests between disk groups such that FC requests never wait on FATA.

[Jeremy C]

I think setting the write cache to "write-through" for FATA vdisks would cause the server writing to a FATA vdisk to be slowed down from writing too fast and therefore save the rest of the EVA from getting backed up behind the FATA requests. Does that make sense?

[Uwe Zessin · Accepted solution]

Yes, in theory. But without wbc, the virtual disks become _awful_ slow.

I had such a case were the customer was too eager to have "cheap storage" and he bought 10 FATAL disk drives for each of his EVAs. Then he was dumping multiple TeraBytes from multiple databases each night.

After some time the problems you mention came up. I was asked for help and did some little analysis with EVAperf and TLVIZ.
When we turned of the WBC, the whole backup did not finish in time and latencies where *horrible*.

To keep the story short:
additional Data Protector licenses were bought to save the databases directly to tape. All FATAL disk drives were thrown out and replaced with smaller FC disks drives. Those were added to an existing disk group for additional storage requirements.

"Friends, don't let your friends use FATAL disk drives!"

[Jeremy C]

I'm ok with the FATA vdisks being _awful_ slow if it guards the EVA from high latency. Seems like the lesser of evils to me.

Thanks for the insightful reply!