Re: buffers vs. pages vs. kernel speed

Dr. Werner Fink (werner@suse.de)
Fri, 13 Jun 1997 11:39:45 +0200


> This patch allows the GFP_BUFFER priority allocations to be successfull
> against all other caches. Indeed it should fix memory lack in some
> situations, however the new balancing sheme may have very weird effect in
> some quite normal situations.
> For example, if you run concurrently reading and writing you probably
> would get the following effects:
>
> - CPU wasting.

With Matthias state fix this should not happen that often.

> - Swapping.

Maybe there is a better solution to avoid unnecessary page I/O ...
the best solution is to add ram :-)

> - Break read-ahead.
>
> In my opinion, if a so simple solution would fix *really* the problem, it
> would have been discovered and applied since months.

The meta problem is: many of the developers have much more ram as the most
high loaded servers out there ... some of them do not need any swap space ;^)

>
> Just try to untar some large files from disk to disk and let us know what
> happens (Or try some benchmarks like Bonnie ) ...

Sorry ... a rock stabel 2.0.31 is needed. Maybe with the buffer code
out of 2.0.29 _PLUS_ the fixes applied in pre-patch-2.0.31-2 to track
the bugs in 2.0.29 down.

Werner