topic Re: High Qlen on a VA7400 in Disk Enclosures

High Qlen on a VA7400

Domenico Viggiani — Tue, 28 May 2002 09:01:57 GMT

Hi,
using Glance, I can see a costantly high value of Qlen for a specific VA7400 LUN (other LUNs of same array are OK).
Where can I indagate furtherly?

Thanks in advance
Mimmus

Re: High Qlen on a VA7400

Paul Barmettler — Tue, 28 May 2002 09:22:56 GMT

Hi Domenico,

is the concerned LUN on the same node like the other LUNS you talk about?

Re: High Qlen on a VA7400

Domenico Viggiani — Tue, 28 May 2002 12:51:24 GMT

Hi,
the LUN of this array are all
mounted on the same system.
Now I tried to set queue_depth to an higher value (using sysctl) and to use primary/alternate path according to some performance tips from HP (primary path on controller 1 for LUN in Redundancy Group 1 and viceversa).
I hope in better performance.

Re: High Qlen on a VA7400

Vincent Fleming — Tue, 28 May 2002 22:31:02 GMT

If you queue depth does not increase when you tune the kernel to allow it, then the disk array is keeping up with the demand. The VA can handle about 256 queue tags, so go ahead and make the queue depth limit 100 and see what happens.

If your queue depth goes to 100 and stays there, you have a performance bottleneck on that LUN. If it does not increase the Glance Qlen, then the array is doing well.

I would say that if it's over 10 to 15, you should split that LUN into two, one on each Redundancy Group (RG), and then stripe them. This should give better performance.

Good luck!

Re: High Qlen on a VA7400

Don Capps_1 — Wed, 29 May 2002 11:52:46 GMT

All good suggestions but perhaps a few questions might be useful before jumping too far. Here are a few that might be revealing :-)

Have you allocated more than 1/2 of the space of the array to LUNs ? and if so, is the array in AutoRAID ?
How many spindles are in the array ?
Are you in RAID 5 mode ? and if so, is your application write intensive ?
The cache segment size on the VA7400 is 64 Kbytes. What is the typical read/write transfer size of the application ?
Did you use vxtunefs on the filesystem to set the value of max_buf_data_size so that it would be a better match for the cache segment size ?
Have you checked the fibre channels for errors using fcmsutil ? Sometimes silent errors can create silent retries and slow things down.
How much memory is in the system ? How much of it is being used for buffer cache ?
(If one satisfies more ops out of buffer cache then the array's queue depth will be small)

Enjoy,

Re: High Qlen on a VA7400

Domenico Viggiani — Thu, 30 May 2002 06:34:12 GMT

Setting controller 1 as primary path for LUNs in Redundancy Group 1 (and controller 2 as primary path for LUNs in Redundancy Group 2) seems to alleviate the problem.

Thanks for all suggestions