Peak Utilisation

[Rushank]

I want to identify the disks or mountpoints being heavily utilised. I would like to analyze the data for last 15 days and look for the those heavily used disks.

I tried doing it by perfview and extract command. But the report does not give me the which disk is badly hit
What is the best way to find out this?

[Christopher McCray_1]

When you used your template file, did you use the variables BYDSK_PHYS_IO or BYDSK_FS_READ
(basically look at the BYDSK variables)

This may be what you're looking for

Hope this helps
Chris

[James R. Ferguson]

Hi:

You might also consider using 'sar -d'. Collect and extact the data you want.

Regards!

...JRF...

[Christopher McCray_1]

The template file for extract is
/var/opt/perf/reptall

and the users guide is at:
/opt/perf/paperdocs/mwa/mwausers.pdf

Hope this helps
Chris

[Rushank]

Hi Chris,

If you are talking about reptall file, then yes I uncommented that out but output doesn't show this..

[S.K. Chan]

This might not be the best way but it's quick and dirty at least. For eg:

# sar -d 15 2 > output.file

The above takes 2 samples every 15 secs.

[Christopher McCray_1]

Could you please post the command you used for extract?

Thanks
Chris

[Rushank]

This is the command I used.

/opt/perf/bin/extract -xp -b today-15 -r /var/opt/perf/reptall -G -f /var/opt/perf/outputfile

| |Num | Phys | Phys | Peak |
Date |CPU % |Disk| IO Rt | IOs |Disk %|
01/15/02| 19.61| 64| 1203.0| 4325927| 75.95|
01/15/02| 22.30| 64| 1499.1| 5392276| 75.63|
01/15/02| 21.28| 64| 1151.3| 4137649| 74.88|
01/15/02| 15.53| 64| 1132.0| 4071889| | 65.55

[Christopher McCray_1]

Replace the -G with a -d (disk device detail) or -D (disk device summary). If you have the BYDSK collectors uncommented, this should give you what you need.

Hope this helps

Chris

[Rushank]

Chris,

This is not giving the output I want..

I want the list of disks or mountpoints , and its peak utilisation for last 15 days

I've sar data for last week but I wanted mwa data so i can easily import into excel.

[Christopher McCray_1]

I don't know haow useful this may be, but let's give it a shot:

There are two metrics you can use, BYDSK_DIRNAME and LV_DIRNAME

the BYDSK_DIRNAME give the name of the file system on a given physical device. This however, may not be useful if you have more than one file system on the disk device, in which case it merely says "Multiple FS"

the LV_DIRNAME may be more useful (haven't tried yet, in where it displays the path name of the logical volume or volume group. This one, in conjuntion with using the LV_* metrics may get you closer to what you need ( the -z and -Z options.

I hope I am getting closer into the neighborhood here.

Chris

[Christopher McCray_1]

Also just noticed something, when you use the -b option, you need to specify the beginning date, meaning the date 15 days ago in mm/dd/yy hh:mm format, followed by the -e where you can then use "today".


What do you think

Hope this helps
Chris

[Rushank]

OK I tried both , the -z or Z gives me error saying

The 'LVOLUME' requested were not found in datafiles.

With the Disk ID BYDSK_DIR NAME I'm getting H/W path but not the disk name.
Surpringly it gives me multiple FS in FS name Infact I don't have multiple FS on any single disk. Each LV is sitting on separate disk.

[Christopher McCray_1]

Maybe a stupid question, so I apologize in advance. Did you also uncomment a metric, like LV_READ_RATE and/or LV_WRITE_RATE ?

Again, probably a stupid question.

Crossing my fingers
Chris

[Rushank]

Yessss.......

[Christopher McCray_1]

well then, in the mean time I was playing around and ran the following command:

# chown mccrayc:sysadm test1
extract -Z -r /var/opt/perf/reptall -b today-15 -e today -f /home/mccrayc/test -xp


It worked.

I did have to modify a coulple of things like uncomment the date field, and when I ran it a second time, I received an error because I had to specify a different file. I am attaching the output, in excel format. I know the numbers are kind of small, but they may be different on your machine. The problem is< I don't know everything you have uncommented, etc. In any case, here is my output.

Hope this helps
Chris

[Rushank]

No luck with LV...

[Dennis J Robinson]

Rushank,

This is going to be easiest with sar.

First make sure sar historical logging is turned on. I do this via a line in root's crontab such as:

0 * * * * /usr/lbin/sa/sa1 600 6

You are able to do a :

sar -d -f /var/adm/sa/saXX

where XX = two digit date within last thirty days.

Do sort on the column with percent busy and you will be able to determine which disks or disks on avg has highest pct busies.

Keep in mind, in PV when looking at the "DISK" section that the DISK BUSY % represents to top % busy of the busiest disk in the system at any given time.

As well as PHYSICAL BYTE rate represents PHYSICAL BYTE rate of the highest throughput disk at any given time...

Dennis

[Rushank]

Hello,

I don't want to do this report with Sar. I've already done this excercise with sar. I am looking something with mwa data.

[Doug Grumann]

There are two kinds of "disk utilization": How busy a disk is processing requests (BYDSK_UTIL) and how full it is, space-wise: LV_SPACE_UTIL. Depending on which you are concerned with, you'd either want to look at the by-disk or by-logical volume data. Note that in C.03.50 (December 2001 release) the by-filesystem class was added as an option but you'd need to turn it on. See the ReleaseNotes for more information.

In PerfView, I use the Class Compare graphs to get a quick look at different disk/lvols. Select your system, select Class Compare, click on Draw, click on DISK, click on Select All, click OK, click BYDSK_UTIL, click Draw, then adjust the graph to your liking with the various Settings.