MSA1000 Disk performance

[Blair Cassidy]

Is there anyway to monitor the performance of the disk in an MSA1000 (ideally real time). We are experiencing queued back logs with our email environment and are trying to isolate where the bottleneck is (disk, fabric, communications,etc) The only thing I see is the ADU utility, But doesn't that really look at hardware failure more so then performance ? Insight manager reports no issues (both server and disk array)
Thanks;

[SAKET_5]

Blair,

Although the MSA1000 performance utiliy is not flash in any way, you could try this to get a realtime view of the disk subsystem performance:

1. from the msa1000 cli, type "start perf"
2. from the msa1000 cli, type "show perf"
3. after you r done with the monitoring, you can stop perf capture by typing "stop perf".

Hope this helps.

[Glenn Strafford]

If I remember right 'Show Perf' from the CLI shows errant readings for Avg Queue depth. I think 'show perf logical' will give you a better understanding of how your LUNs are performing. The Avg write and read latencies are the last 2 columns in that output. The numbers are in milliseconds. If I remember right one of the SAN gurus on here said 25ms and under is good.

If you are running Win2003 make sure write caching is enabled in windows. To do this go to the properties of the disk in windows that is on the SAN LUN. On the next window click the hardware tab. this should show a listing of all hard disk drives. One of them should be listed as something like MSA 1000 Volume multi path disk device. Select it and click the properties button. On the next screen go to the Policies tab and ensure that the following are selected: Optimize for Performance, Enable Write Cache on Disk, Enable Advance Performance.

[Glenn Strafford]

I forgot to mention upgrading to latest firmware may also help you with performance. 5.20 gave me some performance improvements over the firmware that shipped with the unit version 4.48.

[Blair Cassidy]

Thanks for the input.
I upgraded the following:
MSA1000 controller from ver 4.42 to 5.20
QL2300.HAM QLogic QL2300 HAM Driver from 6.80.08 to QL2X00.HAM Version 6.90.13 (m)

A slight improvement in performance, But not good enough. Since the upgrade I can no longer see the MSA1000 under Mass Storage in Insight manager (just port 2301 on a server running the insight manger agents on the SAN). I use to be able to see this. The reason I bring this up is I have acquired another 256 Mb cache for the MSA Controller and I wanted to verify I am currently running at 256 and have capacity for the second cache module.

Also show perf logical from the CLI shows the SAN environment is more read intensive then write (see attched). I have the read/write at 50/50. I want to change this to read/write at 70/30 respectfully. I see in the docs how to do this but can this done dynamically without a system outage and if so does the SAN take a performance hit while the disk is reconfigured
As always thanks very much for the assistance
Blair

[John Kufrovich]

Blair

You sure have some latency on those drives. Can you provide a >show tech_support from the cli.

You can see if you are utilizing the cache on the MSA.

>show cacheinfo
This is a snapshot. Cycle through it a few times.

jk

[Blair Cassidy]

Here is the file.
Thanks for the assistance

[Blair Cassidy]

In the file under Cache: what does
Batteries are disabled and
Module #1 is fully charged and turned off.
mean
doesn't sound good But not sure
Thanks

[John Kufrovich]

It means you batteries are fully charged and we didn't need to turn on BBWC to save something important.

Which is your mail server?

From that one cacheinfo the MSA isn't busy.

Issue a >start perf and let it run for a while. This will allow all the numbers to settle. Then provide >show perf logical
If your latencies are above 25ms. You should consider one or all of the following.

1). Add more spindles. It doesn't look like a viable option.
2). Increase your stripe size to 64K.
3). You may have to move your LUNs around.

One item you need to consider, the more LUNs you carve out of a disk array, the more the access pattern to those LUNs appears random. It takes time to seek those heads between the different LUNs.

[Blair Cassidy]

Which is your mail server?

The problem environment (the one taking the biggest hit AveRd Lat is 180ms to 280ms)

Parallel SCSI Array C with Spare
A four node cluster (file server FS1810, FS1811, FS1812, FS1813) with each server runs a post office on each LUN)
Disks are the following

LUN 15 (ACU view 16) - just 20 Mb for operational Cluster space at the OS level
LUN 16 (ACU view 17)
LUN 17 (ACU view 18)
LUN 18 (ACU view 19)
LUN 19 (ACU view 20)

2). Increase your stripe size to 64K.

So if migrating a post office (LUN 19) out of this environment (to a standalone separate server NOT running on SAN disks). Using that space to expand the array , Recreate LUN from 120 Gb to 100 Gb. Then increase stripping to 64 and downgrading to RAID 5 (instead of 6- ADG).

a) Would I have the â netâ disk space to accommodate the new striping from 16K to 64k (20 Gb plus moving from raid 6 to raid 5).

b) It looks like the rebuild would be 25hrs/ 100 Gb. I realize we would take a performance hit but would the mail system still function ?

3). You may have to move your LUNs around.
Not sure what you mean as there is a basically 4 LUNS mounted one / server and only these 4 servers can see the 4 LUNS (SSP) in the array. If I were to move the LUNS what would I base the move on ?

Finally my biggest hits are read So would moving from 50/50 to 70/30 help at all ?

Thanks again for this help
Blair