performance problem with va7400 v16

[Saleem_3]

Well hi guys,

I am experiencing lot of performance problem with the va7400 even though firmware 16 is installed.Hp was unable to figure out why.
i have two brocade switches and one va7400 of 10 disks 36gb.the performance i got from commanview is below 20mb/s which is inacceptable.Can you help me please

[Vincent Fleming]

Can you please explain what you are doing with the array (database, streaming video, file serving, etc), what kind of host(s) you are using, and your LUN layout. (number and size of luns, etc)?

20mb/s is not so bad, if you're using a database... how many iops are you getting? what's your block size?

[Eugeny Brychkov]

In addition here are some tips:
- did you setup VA to work in fabric?
- did you setup VA with correct host behavior table (or default VA host behavior) and security table (if you have security turned on)?
- do you have latest drivers for FC HBAs in the server(s) and do they work in fabric?

- which command(s) do you use for (emulating) workload? You should know that some hpux commands could not be used for perfomance measurement
Eugeny

[Don Capps]

From my humble experience:
If you are performing sequential I/O testing then you may wish to use the command "vxtunefs" and set the read_nstream and write_nstream to a larger value. Also, the value of discovered_direct_iosz may be causing the buffer cache bypass to become active. This would depend on your applications transfer size.
Also, the VA7400 uses an internal 64 Kbyte segment size in its cache structure. So... it helps to tell vxtunefs to set the preferred_io_size to be 64k.
I also increased the depth of the SCSI command tag queue in the system file. This helps get more parallelized activity going on the array.
Again, all of these tunes are for a sequential workload. On my system it altered the performance from 20 MBytes/sec to around 80 Mbytes/sec. (limited by my FC controller).
Enjoy..

[Roger_22]

Humble experience, my ass. Don knows!

Do this. 1) Upgrade to HP17 and Command View 1.05. It has an improvement to the write cache algorithm. Also, enable the new sequential pre-fetch algorithm.
2) Check the eventlogs. If you have a lot of entries during a period you are having poor performance, then the array might be broken, have HP look at the logs.
3) 10 disks is not much. Learn how to read and interpret the disk performance logs. Check you average transfer size and the balance between the two RG. If they are not what you expect, then you???ll need to work on your application/OS settings.