Miss read KB/s value on EVA Perfmon

[JorisK]

Hi,

I'm investigating why our EVA4400 SAN locks up a lot, usually always during the backup.

I'm currently running perfmon to search for strange value's. I have to say, i don't understand what most value's exactly mean but i just give it a shot.

I'm currently watching the value Read Miss KB/s on my virtual disks, and i'm monitoring all instances possible (which are 12).
During daytime/production not much happens here but halfaway during the backup one of the instances jumps to value's between 2000 and 50000.

What does this value mean and how can i identify the instance to a physical piece of hardware? Could these value's mean faulty hardware?

Regards,
Joris

[Armin Kerl]

Hi,

"Miss" is the Value for Data that is not comming from the Cache (Cache Miss).

[JorisK]

Okay, would this be a problem?

Could this indicate my cache size is too small?

[Armin Kerl]

I think this is normal, if you Stress the EVA with Backup.
You can not Size the Cache.

We dont know how many Disk (10k/15k) you have, what your Tape Device is and also what your Problem realy is.

But your EVA is able to deliver Up up 50MB/s (or may be more), this is O.K..

Download "HP Library & Tape Tools", go to Performance, generate a 5GB Load (4-5-272Files, I Think).
Do a Pre-Backup and than Restore Test to a LUN and see what MB/s are able.

Compare this with your Tape Devices.

[JorisK]

I'm not using our EVA as the backup destination, but as source.

A backup server reads out all info on the EVA and stores it on the backup server which has local attached storage.

So we're talking about read I/O only here...

[Armin Kerl]

First:
Doing a Backup: We're talking about read MB/s only here and not read I/O's, this is a different.

Second:
The LT&T Perf Test will give you the Read MB/s from the EVA LUN.

Third:
If you don't like to give us the requested Infomation (Disk, Problem) and don't like to Test as recommended, I will end this communication.

[Víctor Cespón]

If you have a high number of non-cached reads, the limit will be the access time of the disks.

If you're doing backup of filesystems, and the files are read in alphabetical order, folders and then files, the disks heads will be jumping all over the place, as the location of the files on the disk platters has no relation to their name.

You need to see how many "Total Read Miss Req/s" you have, and divide that by the number of disks on the disk group.