still problems with heavy i/o load

still problems with heavy i/o load

[Roy Sigurd Karlsbakk]

hi

I don't know what this might be, but still, now on -rmap12a, i get the
following behaviour:

* streaming starts
* speed is initially >40MB/s
* when cache is used up, it falls to ~30MB/s - then (after a while) down
  to ~25MB/s
* then down to 0, which might show the wget processes on the remote
  computer should be finished, but they aren't. They (59 of the original
  100) are in Sleeping state. The server won't push more data.

This problem is _not_ rmap specific, as mentioned in
http://karlsbakk.net/dev/kernel/vm-fsckup.txt. With 2.4.17-vanilla, the
data transfer halts after reading 2xRAM bytes.

strangely, rmap11c seems to be quite stable, but only gives me ~32MB/s,
whereas the initial is close to 50.

I have posted mesages about this bug so many times now, that I really soon
will try to install CP/M or something. At least a stable system!

And - yes! - I have tried Andrea's patches. The only fscking thing that
seems to be close to solving it is rmap11c

Please help me about this

--
Roy Sigurd Karlsbakk, MCSE, MCNE, CLS, LCA

Computers are like air conditioners.
They stop working when you open Windows.

Re: still problems with heavy i/o load

[Roy Sigurd Karlsbakk]

> strangely, rmap11c seems to be quite stable, but only gives me ~32MB/s,
> whereas the initial is close to 50.

This is not the case. After some time it failes on the same point, leaving
60 blocked clients on the client computer.

--
Roy Sigurd Karlsbakk, MCSE, MCNE, CLS, LCA

Computers are like air conditioners.
They stop working when you open Windows.

Re: still problems with heavy i/o load

[William Lee Irwin III]

On Wed, Jan 30, 2002 at 06:42:38PM +0100, Roy Sigurd Karlsbakk wrote:
> I don't know what this might be, but still, now on -rmap12a, i get the
> following behaviour:
> * streaming starts
> * speed is initially >40MB/s
> * when cache is used up, it falls to ~30MB/s - then (after a while) down
>   to ~25MB/s
> * then down to 0, which might show the wget processes on the remote
>   computer should be finished, but they aren't. They (59 of the original
>   100) are in Sleeping state. The server won't push more data.
> This problem is _not_ rmap specific, as mentioned in
> http://karlsbakk.net/dev/kernel/vm-fsckup.txt. With 2.4.17-vanilla, the
> data transfer halts after reading 2xRAM bytes.

This is very strange. Is the client machine constant? What kernel does
it use? Is it reproducible against multiple client kernels? This sounds
like a fairly serious regression. Are you always using tux as the httpd?
What about other httpd's?

On Wed, Jan 30, 2002 at 06:42:38PM +0100, Roy Sigurd Karlsbakk wrote:
> strangely, rmap11c seems to be quite stable, but only gives me ~32MB/s,
> whereas the initial is close to 50.
> I have posted mesages about this bug so many times now, that I really soon
> will try to install CP/M or something. At least a stable system!
> And - yes! - I have tried Andrea's patches. The only fscking thing that
> seems to be close to solving it is rmap11c
> Please help me about this

I will at least attempt to reproduce this behavior. I'm suspicious that
the problem could lie in a dark corner only tangentially VM-related, as
you seem to be able to reproduce it under a variety of VM's.


Cheers,
Bill

Re: still problems with heavy i/o load

[Roy Sigurd Karlsbakk]

> This is very strange. Is the client machine constant? What kernel does
> it use? Is it reproducible against multiple client kernels? This sounds
> like a fairly serious regression. Are you always using tux as the httpd?
> What about other httpd's?

It's not related to the client at all. It's reproducable with

dd if=file00 of=/dev/null &
dd if=file01 of=/dev/null &
dd if=file02 of=/dev/null &
dd if=file03 of=/dev/null &
dd if=file04 of=/dev/null &
.
.
.
dd if=file99 of=/dev/null &


It's reproducable on all 2.4.x I've tried. That is - with slight
differences
>
> On Wed, Jan 30, 2002 at 06:42:38PM +0100, Roy Sigurd Karlsbakk wrote:
> > strangely, rmap11c seems to be quite stable, but only gives me ~32MB/s,
> > whereas the initial is close to 50.
> > I have posted mesages about this bug so many times now, that I really soon
> > will try to install CP/M or something. At least a stable system!
> > And - yes! - I have tried Andrea's patches. The only fscking thing that
> > seems to be close to solving it is rmap11c
> > Please help me about this
>
> I will at least attempt to reproduce this behavior. I'm suspicious that
> the problem could lie in a dark corner only tangentially VM-related, as
> you seem to be able to reproduce it under a variety of VM's.
>
>
> Cheers,
> Bill
>

--
Roy Sigurd Karlsbakk, MCSE, MCNE, CLS, LCA

Computers are like air conditioners.
They stop working when you open Windows.