UNIX for Dummies Questions & Answers

Hello,

The last line of prstat shows load average.
I am unable to figure out what actually it is.
I have read the man pages and also googled, all for no use.

Can somebody help me, as to what should be the avg. load of the system for best performance and how is this load of prstat calculated.

=)
It is the same as the load average when you run a 'uptime'

Do a man page on uptime:
'run queue over the last 1, 5 and 15 minutes.'

Performance is subjective...
You talking about throughout or latency or response time or bottleneck or etc..... They are all performance..

If you are looking for a magic number, I supposed for UltraSPARC II, Adrian cockroft from SUN mentioned preferably less than 4~5 runnable jobs per CPUs, (So you see, it depends on the quality and quantity of CPUs you have)
to avoid spending too much time in context-switching.
This load average is the same as the 'r' you get from vmstat. Some people call this CPU queue length.

If you are asking about this to solve a performance problem, you need a baseline to compare.

E.g.
I have a Oracle DataMart and Warehouse of 6 CPUs (UltraSPARC III), often hover around 40 jobs. No user complain.
I also have another performance sensitive Oracle OLTP DB with 6 CPUs (UltraSPARC III). Once the queue hits 20, user complain.

Aha,

That means this load avg. is the average no. of process in the last 1, 5 and 15 mins.

It says run queue, that means its irrespective of the no. of CPU or will it depend?

Despite the documentation, the load average is not computed by looking at the run queues. Instead, the process table is scanned. A process is counted if:
1. it is running
2. it is ready to run (on a run queue)
3. sleeping on a high priority event

Number three usually means that the process is waiting for disk i/o to complete.

So the load needs to be compared against the number of cpu's. A single cpu system is very busy at a load of 7. But an 8 cpu system would be not very busyl at a load of 7. And it takes more to drive an 8 cpu system up to a load 7 that it takes to drive a 1 cpu system up to a load of 7. So the number of cpu's is very important.

Thanks,

Now comes the main part, I think i have got a hint on that in the earlier post, but here is a twist.

Can i somehow find out a suitable one.
By response time to a particular task or something. Because i have to find the best load for my system as i have made some scripts which will utilise the whole system for about 12 hours, i need to customise it based on load avg. so that other users don't get any trouble.

Just lower the priority of your program. Then let kernel decide when to run your program. See "man nice".

I can't lower the priority as its of topmost priority.

I just wanted to get an balance situation.