Re: [Openipmi-developer] IPMI driver performance problem

[Nathan S. <ns...@ac...>]

On Mon, 2008-03-17 at 20:21 -0500, Corey Minyard wrote:
> 
> This shouldn't significantly affect performance.  kipmid runs at nice 
> level 19, meaning that basically everything else is more important
> than 
> it.  It could be made a little bit lower priority by setting the 
> scheduling policy to SCHED_IDLE, but that wouldn't make a big
> difference.
> 
> There's a difference between running and taking away performance.  It 
> shouldn't be taking away performance.  It may be using CPU, but only 
> when the other processes are idle, unless you have nice-ed them to
> lower 
> priorities (bigger nice numbers).
> 

Thats a good point, it hadn't occurred to me to look at nice
level.

One of the other problems I have is that I'm looking for other,
real kernel performance problems (the machine has several other
scalability issues, unrelated to IPMI, but often in the kernel).
And so having IPMI come in and consume large amounts of time in
the kernel makes detecting these other issues more difficult.

I think I can resolve this though, by keeping closer track of
when the userspace code that makes IPMI driver calls is running
and suppress warnings that would fire off during that time.

thanks.

-- 
Nathan