Re: Very bad interactivity with 2.6.0 and SCSI disks (aic7xxx)

[Nick Piggin]

Stefan Foerster wrote:

> [I've sent this mail aleready, but got an error from my MAILER-DAEMON.
> Perhaps ist was too large vor lkml, so I'm moving the oprofile output
> to a webserver]
>
> * Andrew Morton <akpm@xxxxxxxx> wrote:
>
> > Stefan Foerster <stefan@xxxxxxxxxxxxxxxxxx> wrote:
> >
> > > * Andrew Morton <akpm@xxxxxxxx> wrote:
> > >
> > > > Stefan Foerster <stefan@xxxxxxxxxxxxxxxxxx> wrote:
> > > > A kernel profile would be needed to diagnose this.  You could use
> > > > readprofile, but as it may be an interrupt problem, the NMI-based oprofile
> > > > output would be better.
> > > Is this procedure documented anywhere?
>
> [every information I needed]
>
> I did the following steps:
>
>
> opcontrol --setup --vmlinux=/usr/src/linux/vmlinux --event=RETIRED_INSNS:100000:0:1:1
>
> Then I used your shell source:
>
> ~/shells/oprofileit dd if=/dev/zero of=test bs=1024 count=1048576
>
> opreport -l  /usr/src/linux/vmlinux  > /tmp/1
> opreport -ld -D /usr/src/linux/vmlinux  > /tmp/2
>
> During the dd, again the xmms playing a file from an tmpfs froze and
> even screen redrawing was very, very slow.
>
> The output of these commands kan be found at:
>
> http://home.in.tum.de/foerstes/oprofile-1
> http://home.in.tum.de/foerstes/oprofile-2
>
> Is this information useful in debugging my problem, or should I go and
> try again with readprofile or other tools?

Nothing jumps out at me. Is your machine swapping during the bad behaviour?
What is the effect of echo 0 > /proc/sys/vm/swappiness, swapoff -a, or
using the deadline IO scheduler (boot with argument elevator=deadline)



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