want to drill down further to find source of disk bottleneck

[Doug Kratky]

With Glance, I've determined a disk and a specific filesystem to be a bottleneck. Now, I'd like to find out even further which specific file or files are the problem. I can use glance or lsof to see which processes have which files open. (It turns out I have close to a hundred processes which are using that filesystem.) But does anyone know of a way to find out which files are getting hit the hardest? Or, given a process, break out its I/O rates by disk or filesystem instead of just a total I/O rate for that process?


Thanks,
Doug

[steven Burgess_2]

Hi Doug

I can be quite hard to catch a process within glance as it may have come and gone in a short period of time.

You could have 2 screens open, the first looking at I/O by file system, the 2nd with the command ready to catch the pid outputting it to a file. With this method you can quickly gather file information from the PID in the NODE column.From the files you could extract the 8th and 9th column (i think) for the mount point and inode

more file | awk '{print $8 " " $9}' >> somefile

then

cat somefile |
while read FILE INUM
do
find $FILE -inum $INUM -xdev -print >> files.out
done

The above will give you the files accessed by the pids.

Otherwise load the trial copy of measureware and perfview from the OS cd's , you'll get 30 days out of it

HTH

Steve

[John Palmer]

Hi,

Not an easy one... With glance you can identify which processes are doing the most I/O.

You can select each process and examine the files it has open with the F screen. Unfortunately glance doesn't list individual file I/O but by checking the Offset value you might get some idea which files have a lot of activity.

Regards,
John

[Rory R Hammond]

find /filesystem -exec fuser -u {} \;


You might have to catch the stderr output.

I do the following
script
find /filesystem -exec fuser -u {} \;
exit
pg typescript

Rory