Which process consume the disk ?

[沈國維]

Hi Pals,

I have a problem with huge disk usege.

When system slow/down , try to use
"sar -d" to verify disk usege.

And find a device almost 100 %busy .

The problem is how to verify which process took large disk resource ?

Wilson

[Muthukumar_5]

You can find memory usage with top or ps commands. To find disk usage, normal os is supporting to find it out. May be glance support?

-Muthu

[Arunvijai_4]

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=744917

-Arun

[RAC_1]

Only glance/pv will help.
glance -i (file systemwise usage)
glance -d
glance -g (at the the end of the screen, it will show process with top disk usage)

[Joseph Loo]

hi,

u may need to refer to this as well:

http://www1.itrc.hp.com/service/cki/search.do?searchString=KBRC00000947&docType=Security&docType=Patch&docType=EngineerNotes&docType=BugReports&docType=Hardware&docType=ReferenceMaterials&docType=ThirdParty&searchCrit=exactphrase&search.x=12&admit=-682735245+1130479645443+28353475&category=c0&mode=id&search.y=10

regards.

[沈國維]

Is any method without glance ?

[Bharat Katkar · Accepted solution]

Disk Usage 100% may not be Disk Bottleneck though it look so.
You might want to monitor memory,CPU as well.

Attached is the performance cookbook for the same.

Hope that helps.
Regards,

[Bharat Katkar]

Hi
One more for you...
Regards,

[沈國維]

As we know :
sar -- system activity report is an good tool to monitor performance.

sar -w -> monitor swapin/out activity
sar -d -> monitor disk activity

[Ted Buis]

You should first check if your disk I/O is to to page-out activity. The "swapinfo -tam" or other commands should allow you to see if there are page-outs. If you are paging out then you want to check to see if your dynamic buffer cache is set too high as it almost always is by default today with larger memory configurations. That can free up memory to be better used by the application. If you are using Oracle, then OnLineJFS can help to eliminate double buffering with its added mount options. If all has been done there and it is real application I/O that is the bottleneck, then you have to focus on spreading out the I/O across multiple spindles. For truely random I/O, disk array cache and buffer cache aren't going to help much, so the number of spindles is very important. You will only get 100 to 200 random I/Os per second per physical spindle so you can look at how many IOPS your are getting and estimate how many disks need to be involved to get you where you need to be. If you are doing significant I/Os to tmp space, you might use mount options on those directories that are more aggressive and faster. The forum might be able to give you better advice with more information on your system.

[Hein van den Heuvel]

Wilson,

100% IO busy might not be a problem.
See for example the classic "HP-UX tuning and performance" book by Sauers and Weygant:

"100% utilization does not mean that more I/Os are not possible. 100% utilization means that each time the disk queue length is inspected, there are one or more I/Os in the queue."
Notw if your disk is really a raid5 set backed by 8 drives, then it is only normal to have 2 - 6 disks with an active IO, so it will appear 100% busy very soon.
To know whether it actually is a problem you'd need to understand the underlying disks: how many, roughly how many IO/sec can it support, roughly how many MB/sec can it support, and what is the actual IO/sec and MB/sec load.
Is this per chance an oracle applications? If so then I find the Oracle statspack to be the best source to describe which IOs and why.
Since you mention slowdown, one assumes the problem is not just 'visual: 100%' but there is somethign really wrong. The other replies give good starting points: make sure it is realy data io, and not paging/swapping. Is there enough memory? Is there some CPU left when the slowdown is happening?
You may also want to look at the TOP utility to point to suspect processes.

hth,
Hein.

[Sebastian Cesario_1]

Try checking with top or ps, appart frm that you must know if there is enough memory on your system, otherwise OS possibly swap in and out all the time.

Regards,
Sebastian Cesario

[沈國維]

We have checked swapping .
If there is no swapin/out , how can we check which processes consume disk ?

[Sudeesh]

One option i could see is to use lsof.

You mentioned few disks are 100% busy, findout to which FS these disks belong. Then use lsof to find out the process accessing these file system.

fuser -cu /FS will also list the process IDs accessing a particular FS.

Sudeesh