Performance CLI Tools

[Karthik S S]

Hi,

Could some one guide me on how to use iostat, vmstat, sar tools. I could find no document which explains this in a understandable manner.

Any good links are also welcome

Thanks,
karthik

[T G Manikandan]

check these man pages
#man 1 iostat
#man 1 vmstat
#man 1M sar

[Karthik S S]

Or may be that I didnt have patience to go thru all those output fields :-( ... Is there any simple Doc. on performance management?

-karthik

[melvyn burnard]

you may want to consider purchasing the book at:
http://www.hp.com/hpbooks/prentice/ptr_0131027166.html

[T G Manikandan]

http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000063210674
http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000066259053
http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000068400543
http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000065680095

[Bill Douglass]

http://www.geocities.com/techdirectory/performance.html

has some good intro references.

http://www.informit.com/content/index.asp?product_id=%7B98614A3F-B3EA-4167-BA4E-3C774E48CFE1%7D

is Linux-specific, but contains some good explanations.

Above all, nothing beats actually using the commands and seeing how the levels change under differing loads.

Try running iostat for a few minutes, then, in another window, kick off a disk-intensive task (copy a large file across disks, of use dd if you have /dev/zero on your system). Notice the change in bsp on each disk.

Likewise, try this with vmstat and monitor the amv, free, pi and po values.

sar -d is a better tool than iostat, in that it gives you %busy, avg wait and avg. service times, so you might want to run it with the same test as iostat and compare the outputs.

Ina ddition to learning how to read these tools, you'll get a baseline of your system's performance during typical workloads. This greatly helps you when performance problems crop up, as you will know what things should look like, and can spot abnormalities.

[Karthik S S]

Really good links ...

Thanks,
Karthik

[Bill Hassell]

The classic Unix tools present information but interpreting them requires a lot of background. HP's Glance does a good job in presenting this data in a simpler format and provides online help with the metrics but you'll still need to look at performance tuning documents to figure out needed changes. A program that uses the classic tools (sar, vmstat, iostat, ps) and presents suggestions on what can be changed to improve performance is called SarCheck (www.sarcheck.com). This tool can also provide HTML pages for use in a web server, ideal for management reporting.