measuring disk performance - va7410

[Tim Killinger]

We're having performacne issues recently after adding additional users. I'm researching hardware bottlenecks, and while I see an average disk utilization of 30% (during prim hours) on my va7410, I'm looking for a IO "que length" metric to determine if my disk array may indeed be a limiting factor.

I have Glance plus - any ideas on que lenght metric or any other suggestions on how to view disk IO as a potential bottleneck?

Thanks!

[Sundar_7]

iostat and sar can be useful.

sar -d

Look for avserv vs avwait/busy metrics

Also look for specific disks that seem to be occupied more than the other disks. That usually boils down to the Volume group lay out and alternate links allocation.

By default LVM will only use the primary link for IO and cannot load balance across the alternate links.

If $$$ is not a concern, consider HP secure path that can do dynamic load balancing across the alternate links.

[Michael Tully]

Have you looked at sar/iostat ?

# sar -d 5 5

You could have a look at scsictl which does some neat things with scsi queue depths. The default is 8, but you could increase that parameter quite easily and online. One thing about 'scsictl' is that it looses the setting if you reboot, from what I recall.

Before you do that, how have you currently set up your LV's?
Are they striped? I am a firm believer in the 'SAME' (Stripe and mirror everything) That way you get the same results across LUN's and not really have a hot and smelly one or two.

As far as a metric is concerned in glance, I could not find one specifically for this. although is could be buried in amongst others.

[Sundar_7]

Rather than Glance, I would go for measureware and review the historic data using Perfview.

Measureware collects atleast 30 different disk/IO metrics, so you are sure to get what you are looking for. If you already have measureware installed and runnign in the system, then you have drill down to the details using perfview.

Also perfview is very easy to use and will get you closer to the root-cause sooner than glance.

[Tim Killinger]

if I try to use sar...

icgsil1:/var/adm> sar
sar: Can't open /var/adm/sa/sa11

Also perfview... where do I get it?

(we're still a little new to UX, and we haven't hired a body in years!)

Thanks!

[Michael Tully · Accepted solution]

Try:

# sar -d 5 5
which will give the current information. You could set up sar to collect daily information. See the 'sadc' man page or have a search around the forums, where there should be some examples.

Glance does collect some data but will depend on how you have your config file set is to how long it will collect. See the /var/opt/perf/parm file which has a rough explanation. You don't need perfview, besides perfview is not free. There are some reports in /var/opt/perf, but your better off creating your own. Consult the manual, it is not difficult to set these up and grab the data your after.

[Sundar_7]

"Glance does collect some data but will depend on how you have your config file set is to how long it will collect"

hmm..that is a news for me. I always thought glance is just an online performance tool and we can never use it store historic data.

Anybody ?

[Ted Buis]

You can look at the array's performance on the array. I think you can do this from a tab in CommandView, and also from the command line. However, you may need to increase the scsi queue depth in HP-UX for the array using ioctl, as the default is only 8, while the array can handle 240 total. Otherwise the computer is waiting on the array to send it more commands serially, and the array can be waiting on the computer for more tasks that it could handle in parallel. Neither is at 100%.

[Alzhy]

If your machine is 11i and is an Enterprise OS.. then you should have Measureware agents already active.. and chances are it's been collecting useful performance data. If you want to customize it so it collects more data and maintains data up to a set size of log files - simply edit /var/opt/perf/parm - and do a "mwa restart scope". Note that parm file changes also affect Glance or gpm.

Sundar, glance via Adviser can actually collect data via scripts if you do not have Measureware.


But for the task at hand, sar -d and iostat would be your best tools to ascertain if your IO is already lacking...

TO gather stats via sar, simply create /var/adm/sa and enable /usr/lib/sa/sa1 script in root's or sys' crontab (see earlier posts)... Once set, you always have a month's worth of comprehensive sar data ..

[Ted Buis]

http://h200002.www2.hp.com/bc/docs/support/SupportManual/c00218935/c00218935.pdf is the CommandView manual and has information about the performance logs in the array (see page 239). See also armperf on page 160.