Re: Bad 1.3.95 interactive performance (fwd)

Gerard Roudier (groudier@iplus.fr)
Mon, 29 Apr 1996 16:01:01 +0000 (GMT)


On Mon, 29 Apr 1996, Matthias Sattler wrote:

> Hiho
>
> Recently I sent to the linux-kernel mailing list:
>
> >
> > Hiho
> >
> > With 1.3.95 and some kernels before (i don't know exactly when it started)
> > I notice unnecessary heavy swapping activity when the system is up some time.
> > Here comes an example scenario:
> > - boot the computer
> > - login into X (via xdm). This first time everything looks normal.
> > - logout
> > - Let the machine alone for some hours (in my case it is usually > 12h)
> > with nothing running except the usual cronjobs.
> > - login again and be surprised of the time it takes to get the desktop
> > (can be several minutes) and the neverending swapping activity of your
> > harddisc.
> >
> > As long as the working set does not change everything is normal, but even
> > the smallest change results in heavy swapping (while the output of free
> > looks normal).
> >
> >
> > As nobody has written about this before, it seems that I'm the only one
> > having these problems. The only unusual detail of my configuration is that
> > I use the ncrBsd2Linux driver (version 1.8).
> > What can I do to fine out more about the nature of this problem (yes I will
> > try the normal ncr driver...) ?
> >
>
> In fact I don't see the problems with the original ncr driver, so it is
> very likely that the bug is somewhere in the ncrBsd2Linux driver. Is
> there something I can do to find out more about the nature of this problem ?
>
> Matthias

I you mean that ncr Bsd driver has something magic, I agree.
While adapting it to Linux, I tried to stay very closed to the original
driver.

Matthias, a driver does _only_ IO operations that are queued to it.
It is a small piece of code that _must_ do what it is asked to it and
nothing else.

My explanation is the following:
When IO are slow, the cache policy algorithm can reuse pages prior
to deciding to swap processes.

Sorry, but I am not going to decrease ncrBsd2Linux performances in order
to be more king with the page cache.

I am not claiming that ncrB2L has no bugs, and that your problem is not
due to the driver.
However, I use X, gcc, etc.. on 2 machines that I switch off only once a
week and I never saw your problem.
It seems (no reply) that nobody else has had the same problem as yours.

Can you send to me (us) the description of your hardware configuration.

Regards, Gerard.