Re: 2.4.18pre2aa1

> I also couldn't resist to add the dyn_sched because I'm dealing with
> servers that needs to run 2k tasks (with only a few of them running at
> the same time loading cpu), so there the reclaculate loop was evil and
> the change is simple enough to be kind of obviously correct (btw: I also

Something has made a dramatic change in dbench throughput:

2.4.18pre2aa2   Throughput 23.4521 MB/sec (NB=29.3152 MB/sec  234.521 MBit/sec)  64 procs
2.4.17rc2aa2    Throughput 19.6605 MB/sec (NB=24.5756 MB/sec  196.605 MBit/sec)  64 procs
2.4.18pre2      Throughput 12.1986 MB/sec (NB=15.2483 MB/sec  121.986 MBit/sec)  64 procs

2.4.18pre2aa2   Throughput 18.6495 MB/sec (NB=23.3119 MB/sec  186.495 MBit/sec)  128 procs
2.4.17rc2aa2    Throughput 14.1212 MB/sec (NB=17.6515 MB/sec  141.212 MBit/sec)  128 procs
2.4.18pre2      Throughput 8.00551 MB/sec (NB=10.0069 MB/sec  80.0551 MBit/sec)  128 procs

2.4.18pre2aa2   Throughput 9.79641 MB/sec (NB=12.2455 MB/sec  97.9641 MBit/sec)  192 procs
2.4.18pre2      Throughput 8.09211 MB/sec (NB=10.1151 MB/sec  80.9211 MBit/sec)  192 procs
2.4.17rc2aa2    Throughput 5.80232 MB/sec (NB=7.2529 MB/sec   58.0232 MBit/sec)  192 procs

Some more test results for recent kernels are at:
http://home.earthlink.net/~rwhron/kernel/repo.html

> URL:
> 
>   ftp://ftp.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.18pre2aa1.bz2

Based on the diff logs, here is a history of changes going into 2.4.18pre2aa1:
http://home.earthlink.net/~rwhron/kernel/2.4.18pre2aa1.html

-- 
Randy Hron