EVA4400 slow response times

[Toine_1]

Hi,

I have a EVA4400 with two disk groups.
One disk group with 8 FATA disks of 1 Tbyte.
Second disk group of 28 fibre disks of 450 GByte 15 krpm.
All disks have HP06 and EVA firmware is 09006000.

These FATA disks are used in a development environment.
28 Fibre disks are used in a production environment.

From time to time the EVA box has very slow response times. On the connected hosts we get a lot of queueing on the LUNs in the second diskgroup.

It looks like the FATA disks which are in another disk group have a hughe impact on the other disk group with fibre disks (28 X 450 GByte disks). Can it be that the cache is not flushed fast enough to the FATA disks and the second disk group is slowing down.
We don't get any hardware or controller errors from the EVA 4400 box.

Has anyone else has seen this problem?
How can I find out what the cause is?

/Toine

[Víctor Cespón · Accepted solution]

Do not saturate the FATA disk group, it will affect the performance of other disk groups too. The EVA controllers do not multitask, they process I/Os to all disk groups sequentially, so the delays on the FATA disk group affect the time it takes to process an I/O on the FC disk group.

I've seen many EVAperf captures when we had problems with the 1 TB FATA disks with firmware below HP06, it was very clear there.

You can capture some data with:

evaperf all â sz -cont â dur 300 â csv -fo evaperf.csv

And upload it here

[Toine_1]

Thank you,

Does it help to disbale the caching of the FATA disks?

/Toine

[Víctor Cespón]

Yes, you can try that, reserving the cache for the fast disks.
The usual combination of few FATA disks + VRAID 5 + mostly writes yields very slow performace, about 15 IOPS per disk. As I said above, always monitorize the traffic to the FATA disk group and do not saturate it.