Re: Performance issue on dual Athlon MP

[Jens Axboe]

On Thu, Mar 07 2002, Alan Cox wrote:
> > BTW, this doesn't seem like a preemption issue, considering that throughput
> > is very definitely affected as well as latency.
> > 
> > Anyway, please let me know if there's anything I can do, within my
> > constraints, to help.  (As you can guess, though, I don't have any kernel
> > debugging experience.)
> 
> It sounds like the hit you are taking is from highmem and I/O (having to
> copy pages lower into memory so the I/O subsystem can use them). Some of
> that is in the hard to fix for 2.4 category with the x86. There are some
> experimental patches around but they are experimental.

If you are referring to my block-highmem patch, then it's not considered
too experiemental anymore. It's been a long time since any problems have
been reported.

Scott, a small profile run would give us the information needed to tell
whether this is a bounce problem or not. Boot the kernel with profile=2,
then do a readprofile -r ; run problematic stuff ; readprofile >
prof_data

and share the prof_data.

-- 
Jens Axboe