System load

[Shahul]

Hi

Can Somebody tell me what is the suggested HP system's load. Or in normal condition what can be a system's average load?

Best of luck
Shahul

[Stefan Farrelly]

The recommended is 1.0 per cpu. So on a 2 cpu server the average load shouldnt exceed 2.0

If your server is running very heavy then average load should be much less than 1.0 per cpu. Our large servers here (6x550 Mhz CPUs) run around <1.0 which is nice and fast response time for our users, if it did run around 6 that would be the maximum, I would consider it too high for fast response times.

[harry d brown jr]

System load is dependent upon what YOU would consider normal. On an "IDLE" system the load should be below 1.00 (one), but sometimes it ramps up, depending upon what is happening.

I have a few systems that have a "load" of 4 to 10, 10 usually being a peak period. As long as the "load" does not severely degrade performance (what the user sees as acceptable response time), then the server is performing within expectations.

When the "load" average is up over 5 and the response time sucks, then it's time to look into upgrades, which could include fixing the software!

I'd suggest using glance / measureware and perfview to monitor and track a system that is suspect.

live free or die
harry

[Trond Haugen]

As usual with performance the answer is "it depends". You cant take a load average by it self and say if it's normal or not. I have seen systems with load average around 1 that has been perceived as a slow system and systems with load average of 5 that has run fine.

Regards,
Trond

[Shahul]

Ok, Fine, Thanks for the responses. So, U mean to say, Once it crosses 5, then only I should take action about that.

Best of luck
Shahul

[harry d brown jr]

Shahul,

...once it crosses 5, then only I should take action about that.

Well that depends what the response time is. You could have a system with a load average of 20 that has decent response time. It really depends on what the processes are doing. There's no magic threshold that should trigger an event. That's why I suggested to use glance / measureware and perfview to monitor and model your machine on an ongoing basis so that you can track trends that might assist you in getting approval for system upgrades.


live free or die
harry

[Paula J Frazer-Campbell]

Hi

The answer of one per cpu is your best guide,

4 cpus then 4.

I have had a server running at 1 with two cpus and it has been slow.

I have also had a K clas - 6 cpu hit 156 (load) for over 10 minutes and I could still login an bring it under control - mind you the users could not do much.

HTH

Paula

[Paula J Frazer-Campbell]

Hi

Further to my last post.

You should know the applications that your users are running and on a regular basis log in as a user and carry out tasks that they do so that you have a good "feel" of the server.

So when the phone call comes with the words " The system is slow", then you KNOW how it is behaving and if it is slow then you can more easily find the reason.


HTH

Paula

[Bill McNAMARA_1]

I guess my systems are in trouble then..
I often start seeing the mouse stutter on loads of A!.. and I only have one CPU ;)

Bill