I'm using simple two loops in different kernel trees:
while true; do make clean; make MAKE='make -j10'; done
which leads into load upper 30. You can see a great performance upto
load to 25 ... 30+ *and* a brutal break down of that performance
at this point. The system is a PentiumII 400MHz with 32, 64, 128MB
(mem=xxx) and SCSI only. In comparision to 2.0.36 the performance
is *beside of this break down* much better ... that means that only
the performance break down at high load is the real problem.
>
> WF> What's about a simple aging of program page cluster or better of the
> WF> page cache?
>
> We do age pages. The PG_referenced bit. This scheme as far as I can
> tell is more effective at predicting pages we are going to use next
> than any we have used before.
What's about a `PG_recently_swapped_in' bit for pages which arn't found
anymore with the swap cache? This isn't a prediction but a protection
against throwing out the same page in the following cycle.
>
> WF> Increasing the age could be done if and only if the pages
> WF> or page clusters swapped in and the program wasn't able to use its
> WF> time slice. Decreasing the age could be placed in shrink_mmap().
>
> People keep playing with ignoring PG_referenced in shrink_mmap for the swap cache,
> because it doesn't seem terribly important. If you could demonstrate
> this is a problem we can stop ignoring it.
>
> Eric
>
Werner
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/