lkml.org 
[lkml]   [2004]   [Aug]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: Serious Kernel slowdown with HIMEM (4Gig) in 2.6.7
Date
From
Darren,
I ran your test of "ls -laR of /etc >/dev/null" on our system and you
need to gen a kernel without himem and try it. I get around .1 seconds
without himem and a whopping 4.45 with himem enabled.

I also tried without preemption and got the same results. It is strictly
do to setting HIMEM in the kernel.

> Hi Lawrence
>
> On Fri, 13 Aug 2004, Lawrence E. Freil wrote:
>
> > Hello,
> >
> > I'm following the kernel bug reporting format so:
> >
> > 1. Linux 2.6.7 kernel slowdown in directory access with HIMEM on
> >
> > 2. I have discovered an issue with the Linux 2.6.7 kernel when HIMEM is
> > enabled which exhibits itself as a slowdown in directory access regardless
> > of filesystem used. When HIMEM is disabled the performance returns to
> > normal. The test I ran was a simple "/usr/bin/time ls -l" of a directory
> > with 3000 empty files. With HIMEM enabled in the kernel this takes
> > approximately 1.5 seconds. Without HIMEM it takes 0.03 seconds. The
> > time is 100% CPU and no I/O operations are done to disk. "time" reports
> > there are 460 "minor" page faults with zero "major" page faults.
> > I believe the issue here is the mapping of pages between high-mem and
> > lowmem in the kernel paging code. This increase in time for directory
> > accesses doubles to triples times for applications using samba.
> > I have also tested this on another system which had only 512Meg of RAM
> > but with HIMEM set in the kernel and did not experience the problem.
> > I believe it only effects the performance when the paging buffers end
> > up in highmem.
> >
> > 3. Keywords: HIMEM, Performance
> >
> Would you be running these in a gnome-terminal, I remember seeing a thread
> that discussed gnome-terminal problems though a quick search did not turn
> anything up. Here is what I get between a xterm and gnome-terminal.
>
> xterm
> # time ls -lR /etc
> real 0m0.381s
> user 0m0.056s
> sys 0m0.130s
>
> gnome-terminal
> # time ls -lR /etc
> real 0m0.869s
> user 0m0.057s
> sys 0m0.141s
>
> I ran this twice in both teminals and reported the
> second result.
>
> system info
> P4 3.06 HT
> 2.6.7 SMP/SMT/HIGHMEM
> 1GB ram
>
> -------------------------------------------------
> Darren Williams <dsw AT gelato.unsw.edu.au>
> Gelato@UNSW <www.gelato.unsw.edu.au>
> --------------------------------------------------


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 14:05    [W:0.048 / U:2.308 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site