Hi all,

I am trying to monitor the resources on some machines. Unfortunately monitoring load or simply traditional CPU/Memory does not provide accurate values for burst workload. So if I rely on these metrics to accommodate traffic I am ending up in too many losses due to slowness in updating these metrics according to my intensive traffic.

Do you know if there are some low level OS metrics which are more sensitive to changes in traffic(maybe some reported by systat ?) which allows me to react more quickly in multiplexing the traffic ?

Thanks in advance,
Robert

Well, you don't say which tools you've used and how you are using them.
Exactly what do you want to measure?
Most tools have options to specify frequency of polling, but of course the more frequently you do that, the more the measuring affects the numbers measured ... its all a bit quantum

As an example, vmstat http://linux.die.net/man/8/vmstat has a 'delay' param that you can specify.

sar only collects every 10m. quite useless if you need a 2-to-3 minute window into what's going on.

I suggest collectl

Actually, sar has an 'interval' option, that can be specified in seconds
http://linux.die.net/man/1/sar

"by default".
Damn you Chris!

Thank you all guys. It was really helpful.