changelogs for 2.4.17rc2aa2 and 2.2.20aa1

I put together these pages to help others understand what is in Andrea's kernels a little better.

http://home.earthlink.net/~rwhron/kernel/2.4.17rc2aa2.html
http://home.earthlink.net/~rwhron/kernel/2.2.20aa1.html

The pages are created from his patch diff logs.

I've stress tested and benchmarked 2.4.17rc2aa2 a lot and it's been very solid. For a workload that creates a lot of processes 2.4.17rc2aa2 has a definite edge. This is easiest to see in a couple unixbench tests:

                              2.4.17-mjc1  2.4.17rc2aa2   2.5.1-dj10

System Call Overhead 352361.7 362120.5 255809.4 lps
Process Creation 817.5 2037.7 1212.0 lps
Execl Throughput 265.8 458.4 316.5 lps

And lmbench; the highest and lowest results of 3 runs were dropped.

Processor, Processes - times in microseconds - smaller is better
----------------------------------------------------------------
OS fork exec sh
proc proc proc
------------ ---- ---- ----
2.4.17rc2aa2 745 2769 9583
2.5.1-dj10 810 3504 11.K
2.4.17-mjc1 1128 4244 12.K

Good stuff comes from a lot of sources, and I'm hoping some of the tree maintainers will start cherry picking from Andrea's tree too. :)

Note: mjc1 above was configured without preempt, rtsched or lockbreak.

2.4.17rc2aa2 consistently does better at dbench too. More results are at:
http://home.earthlink.net/~rwhron/kernel/repo.html

Randy Hron

Consider it bookmarked 8)
I hope you can find the time to keep up these tests, as they should start to show some even more interesting results as the various trees progress.

| Dave Jones.        http://www.codemonkey.org.uk
| SuSE Labs

I'll second that, very nice that someone takes the time to do this.

Jens Axboe