glance and I/O

[Chris Fadrowski]

While doing some application and Database stress testing today we were viewing glance. It was reporting back DISK/UTIL 100%
I am connected to a SYMETRIX for disk with 4 gb of cache. Just wondering if glance is not reporting correctly. here is the version info

glance C.02.40.00 10/07/99 HP-UX 11

[S.K. Chan]

What does "sar -d" shows ? Just wanted to find out if "sar" and "glance" are consistent or not.

[Michael Tully]

The version your using is quite old. I have 3.35 on most of my systems. How long does the disk IO report 100% usage. Just a moment or the limit for quite a while. The best way to utilise the glance product is to collect the information (scopeux daemon) does this and then you can extract information like for disk IO over a period of time.

What is your database?
Are you using raw logical volumes or mounted filesystems. Sometimes having incorrect mounting options cold cause performance headaches.

[John Bolene]

HP measurement statistics are based on using slow disks that only handle one I/O at a time.

Thay cannot report accurate numbers for disk utilization on fast disks like VA, XP, and EMC.

The fast disks can handle multiple I/O's active at one time. This causes the HP stats to reflect 100% utilization while I/O's are active.

[Martin Johnson]

Older versions of glance have a problem that show disk utilization at 100%, but the disk doing little or no I/O. Upgrade to the latest version.

HTH
Marty

[Sridhar Bhaskarla]

I am not too concerned if the disk util is at 100% on cached disk arrays. Observe the QLEN column in the same window. As long as it is not a high value, it should be alright. Also look at the Phys IO column and see how many IOs are being generated on one interface. Check with your vendor to see if the backend can sustain these many IOs.

Also look at sar 2 20 output and see if the %wio is more than 15.

-Sri

[Chris Fadrowski]

our test is over so i can't use SAR. Tryed that because that is what i usually use but the test was complete already. It showed 100% for the entire time a couple of processes were running (55 minutes). I am pulling stats off of my sym tomorrow to see if it is different however. I do need to upgrade glance but was wondering if i could trust the 100% on glance.

[harry d brown jr]

John Bolene, interesting, I have never heard of that, but good to know!!!

Chris, Because you have a Symmetrix, i'd suggest getting powerpath, work load analizer, basically the ECC suite of products.

live free or die
harry

[John Poff]

Hi Chris,

I have a name for this. I call it, "Glance Busy Disk Syndrome". I usually get it from one of the local DBAs or other wizards who have wandered into Glance, and they start freaking out when the disk utilization hits 100%. :)

The trick here is that Glance is reporting disk I/O for the _busiest_ disk on the system, not for all the disks. It is very possible to have one logical PV that is getting hammered while the rest of the system is fairly quiet. Glance is a great tool but it doesn't understand disk arrays very well. I like to use MeasureWare and look at I/O across filesystems, and the performance of individual processes. If you have MeasureWare installed, you can go back and take a good look at your system even after your stress test is over. Try 'man extract' and search the forum here to get an idea of what you can pull out of it.

Have fun!

JP