Configure cpu interrupts manually

honeybadger · 12-17-2010, 11:27 AM

I was installing freebds for one of my friends and during the entire install the top never crossed 1.0. When I install linux top always shows a cpu load of more than 1.0.
Also on the bsd when copying a huge file (4gb) it takes a long time but then the load never goes beyond 1.0. Linux will do it much faster but then there is a difference in the cpu load time.
What I want to know is is there anyway I can interupt the processor so that the load does not go beyond 1.0.
Thanks for reading.

orgcandman · 12-17-2010, 12:33 PM

Under linux, load != cpu utilization. Under freebsd, it may.
The relationship between the two is indirect. The load average on linux only measures the task runqueue length. See http://www.teamquest.com/resources/g...ay/5/index.htm for more information.

Clear up the situation, a bit? High uptime load averages, under linux, aren't necessarily a bad thing.

honeybadger · 12-17-2010, 12:40 PM

Sorry if I was not very clear in the first post. In bsd I saw that the interrupts were very often and it took the load level down.
Is it possible for me to configure the interupt the processsor when say it is at .80 so that it would not cross .90 atleast. Since linux is so flexible I was wondering if I ever wanted to do that would I be able to.

Sergei Steshenko · 12-17-2010, 02:12 PM

Quote:

Originally Posted by SilverBack

Sorry if I was not very clear in the first post. In bsd I saw that the interrupts were very often and it took the load level down.
Is it possible for me to configure the interupt the processsor when say it is at .80 so that it would not cross .90 atleast. Since linux is so flexible I was wondering if I ever wanted to do that would I be able to.

How about simply using 'nice': http://linux.die.net/man/1/nice ?

honeybadger · 12-17-2010, 02:17 PM

Well I had thought about nice but then it cannot be changed at runtime - unless you are the root. So that is something I would not do.
Just hope I do not have to compile a kernel to get this functionality

.

honeybadger · 12-17-2010, 02:18 PM

Well I had thought about nice but then it cannot be changed at runtime - unless you are the root. So that is something I would not do.
Just hope I do not have to compile a kernel to get this functionality

.

Sergei Steshenko · 12-17-2010, 02:28 PM

Quote:

Originally Posted by SilverBack

Well I had thought about nice but then it cannot be changed at runtime - unless you are the root. So that is something I would not do.
Just hope I do not have to compile a kernel to get this functionality

.

I do not understand what is the problem with load greater than 1 in the first place.

syg00 · 12-17-2010, 05:24 PM

So, you're not prepared to simply use su{do} to adjust a nice value, but you're prepared to hack to kernel to try and achieve this stupidity. You don't understand what you're seeing, but demand an answer to fit your mis-perception.

Go right ahead - it's your system.

honeybadger · 12-18-2010, 12:21 AM

@syg00 - No I am _not_ ready to hack the kernel to get this functionality. Using 'nice' is ok but to be honest I have rarely used it. Using 'nice' to achive the result is fine but then do you realise that I have to use nice _every_ time I am doing something computational intensive.
What I am looking for is somehow a more permanent fix where the load on the cpu does not cross 1.0.
This is not only about my system - I also want learn something more about the way the way I can control the processes eating away at the cpu.

syg00 · 12-18-2010, 01:15 AM

See post #7.
Loadavg is not "load on the cpu". The link above is relevant for Unix (including presumably freebsd) - not Linux.

Go read this. It's not completely accurate these days, but will give you a better idea of what loadavg is. Setting a "watermark" of 1.0 is meaningless - especially on multi-core/hiper-threaded chips, but even on single CPU.

Sergei Steshenko · 12-18-2010, 09:30 AM

Quote:

Originally Posted by SilverBack

@syg00 - No I am _not_ ready to hack the kernel to get this functionality. Using 'nice' is ok but to be honest I have rarely used it. Using 'nice' to achive the result is fine but then do you realise that I have to use nice _every_ time I am doing something computational intensive.
What I am looking for is somehow a more permanent fix where the load on the cpu does not cross 1.0.
This is not only about my system - I also want learn something more about the way the way I can control the processes eating away at the cpu.

There is nothing to fix. Furthermore, if you really want to increase the frequency of HW interrupts, you will be squandering CPU resources - every HW interrupt is an overhead because of saving/restoring registers and the need of kernel to check what to do next, i.e. to check whether current task needs to be suspended and which task needs to be resumed.

orgcandman · 12-20-2010, 12:25 PM

Quote:

Originally Posted by syg00

See post #7.
Loadavg is not "load on the cpu". The link above is relevant for Unix (including presumably freebsd) - not Linux.

Not sure if you're referring to my link, but the link I posted specifically outlines the fact that under linux load avg is NOT the same as task runqueue length. It uses Solaris as a comparison tool to show that under some systems, the two are directly related. Regardless, both above links were quite relevant, unless you can detail why they werent?

Additionally, the first line of the second post:

Quote:

Under linux, load != cpu utilization.

And the last line:

Quote:

High uptime load averages, under linux, aren't necessarily a bad thing.

The OP clearly didn't read that information.