Filesystem Statistics

[Ty Roberts]

I am hoping someone may be able to guide me in the right direction here. I am trying to run some stats reports on a certain filesystem to help our DBA's argue some problems with Oracle. I have GlacePlus and MeasureWare but have not been able to get the information I need. I need to find some command that I can fun that can give me sar -d like statistice but just on 1 filesystem. The filesystem is made up of quite a few disks so I can not just grep for it. Does anyone know a quick and easy way to get some usage / load stats on a particular lvol or filesystem??

Thanks for your help,
Ty

[Peter Kloetgen]

Hi Ty,

do you want the "percent used" value? Then simply use bdf or du command.

Remember, Oracle is using only half of the space for data often!


Allways stay on the bright side of life!

Peter

[Ty Roberts]

Sorry maybe I was not that clear.. I mean like usage / load statistics... Like how many reads, the %busy the disk is.. Basically the type of stats I would get from sar -d

[Tom Geudens]

Hi,
If you are talking "percentages" ... I don't think you can do that on filesystemlevel (I don't think it would make sense either). However, there is an option "IO By filesystem" (option "i") / "IO By logical volume" (option "v") in glance.

Hope those help,
Tom

[RAC_1]

I can recollect the command, but with glance you can get the statstics for a file system-for a logical volume as bellow

disks-reports-io by logcal volumes

Regards,

[steven Burgess_2]

Hi

You can see what is accessing the filesystem with the lsof tool which you can download from the following site

http://hpux.cs.utah.edu/hppd/hpux/Sysadmin/lsof-4.61/

To see read and writes you simply want to map the disk from sar -d to the filesystem with pvdisplay -v /dev/dsk/c*t*d*

HTH

Steve

[V. V. Ravi Kumar_1]

hi,
from command line try
glance -i

regds

[Ty Roberts]

I have glance and I can sit there and watch the disk activity rise and fall, but I need to generate some sort of report to send off site to Oracle. They are basically comming to us saying that the Oracle Log Filesystem is getting hit too hard and that is what is causing our problems. We know this is not the case but need to send them data to back it up. I found a post one where you can use GLancve to generate ascii type of output reports but can not find it again.. It also does not help me to go as granular as the disk level since this filesystem is made up of over 20 disks....

[PIYUSH D. PATEL]

Hi,

For such statistics, Perfview is a better product. Since glance will not store any historical results. I recommend you to go fro perfview. Or else you will have to manage with Glance and sar.

Piyush

[Ty Roberts]

I do have perfview but after fuddling with it for a few hours I could not get the stats I wanted. Can anyone help me wade through the mess of perfview and direct me to a report on disk utilization??? Or even better.. Lvol or FIlesystem Utilization??

[Tom Geudens]

Hi,
I think the post you are looking for is http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xad64c5ea0230d411ade80090279cd0f9,00.html

Regards,
Tom

[Dave Chamberlin]

Do your DBAs know about the utlbstat/utlestat scripts that come with oracle? They collect some read/write statistics at the datafile level in oracle. This can help identify which datafiles are consuming most of the IO on your system.

[Ty Roberts]

Thanks to everyone for their help.. I think I got what I needed to accomlish what I want to... I am not sure if our DBA's know about those scripts.. but I will be sure to let them know about it..

Thanks alot,
Ty

[George A Bodnar]

What version of Measureware do you have? At version 3.50 you can get the following disk statistics per physical disk that should help quite a bit. You can also get KB written/read, etc. but these 3 statistics can show you quickly if you have a disk that is too saturated.

BYDSK_UTIL
BYDSK_REQUEST_QUEUE
BYDSK_AVERAGE_SERVICE_TIME

[MANOJ SRIVASTAVA]

Hi Ty

Essentially what you are looking is statistical report for the disk activity , you may like to user perf view which can store inforation on the local machine and cna be pulled out , drilled etc . This I think is rpiced software and would rqire some licenses , you may alo try to look at sarcheck , this si a free ( one month ) utlity from www.sarcheck.com and is helpful too . I have attached a sample report from one of our systems for yuo refernce.

Manoj Srivastava

[Martha Mueller]

From the graphical version of glance - gpm, I pulled up the menu bar - Reports, Disk Info, IO by filesystem. From there, you can take the menu bar - Configure, and choose the metrics that are meaningful. The one FS_PHYS_IO_RATE might be interesting. That's the number of physical IOs per second directed to this filesystem during the interval ( which is however long you want to watch ).

[Mladen Despic]

Ty,

You can use Glance in "adviser only" mode to generate the reports that you want. You just need to create a file (adviser syntax) where you tell glance what you want. Then run something like this:

glance -adviser_only -syntax -f -j 300

You will need to use "FS LOOP" and/or "LV LOOP" statement to get FS and/or LV types of metrics.

For more information, see documentation under /opt/perf/paperdocs/gp/C/
on your system.

I have also given some examples in regards to a similar request related to memory reports here:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xa56442308663d611abdb0090277a778c,00.html

Mlade

[Frank Slootweg]

> I need to find some command that I can fun
> that can give me sar -d like statistice but
> just on 1 filesystem. The filesystem is made
> up of quite a few disks so I can not just
> grep for it.

Why can't you grep for it?

You have the filesystem name. bdf(1M), mount(1M), etc. give you the Logical Volume name. "lvdisplay -v" gives you the Physical Volumes (/dev/dsk/...). You can then grep(1) the "sar -d" output for those PVs/disks.

[Ty Roberts]

There are 20 disks in the volume group.. Can YOU tell me a way to grep through the output for all 20 disks?? WIthout doing it one at a time???

[Frank Slootweg · Accepted solution]

> There are 20 disks in the volume group.. Can
> YOU tell me a way to grep through the output
> for all 20 disks?? WIthout doing it one at a > time???

"man grep":

> SYNOPSIS
...
> Call with (multiple) -e pattern
...
> Call with -f file

[Ty Roberts]

I ended up just writing a little simple shell script to parse through tht output.. now my problem is trying to convince the DBA's and Oracle that there is no disk contention.. thanks to everyone for their help...