UNIX for Dummies Questions & Answers

Hi,

I got a request from a client to check on their server's cpu, memory and disk utilization. Am planning to use the sar and iostat commands to achieve this. The server is used for sending and receiving messages between the stockbrokers, so I was thinking to run the commands during the market peak period, as this is when loads of messages will be forwarded to the server. This 'market peak' period will last about 10 - 15 minutes.

Question is:

What is the recommended interval/count values should I set for both sar and iostat in order to get a good peak/average utilization result?

Any advice is appreciated.

Thanks in advance.

You would be better using NMON which will create a CSV file which can then be used with NMON analyser to produce an EXCEL spreedsheet file with pictures showing all machine activity. Much easier to use than SAR and IOSTAT and pictures are better evidence to support troubled systems.

Thanks, I've not used nmon before, but I'll try it out.

If you are going for plain text you should log some time before the peak times and after it. Having 5 sec intervals should be ok. I usually just collect in 1 second intervals (depends on how much resources you are looking at - ie. an iostat with a lot of disks will produce a lot of logs).
Though, as johnf said, if you have to present this to other people, that will see nothing in plain text, it's better to use nmon for example.

If you go for plain text, you might think about vmstat to get a 1st impression what is going on.

Code:

vmstat -I -t 1 | tee vmstat.out

The parameters include a timestamp and keep some columns out nobody needs anymore. The 1 is an interval of 1 second and it is piped into "tee", so it will write it to the file and print it on the console too.

Thanks zaxxon. Guess it'll be fine if I were to run it with 5 sec intervals for about a minute before and during the peak time? That'll be 12 counts altogether in a run.

The vmstat flags aren't valid (am running on Ubuntu). Prolly I hafta use a script to append the timestamp in the result.

Thanks again for your input.

Ah, sorry, was the parameters for vmstat on AIX. Have to put my blinders away sometimes
Yep, 5 secs is ok. Let it run some mins before your peak starts and some mins after it so you can see and show others the difference.
But as johnf said, nmon is really good, especially with the easy output in graphs.

Ah.. thanks a lot for the advice!