I'm very happy with 2.4.17rc2aa2. Some things I've run on it:
Laptop PII with 192MB RAM
15 simultaneous copies of crash01 doing 10,000,000 iterations each.
crash01 generates random code and executes it. This ran for over
36 hours and the test completed normally. An earlier 2.4.x
kernel locked up or rebooted on the first iteration. Similarly,
crash02 tests (execute random syscalls) runs for long periods
with no problems. (both of these tests could be dangerous, so
be careful with them).
K6-2/500 with 384MB RAM.
Several executions of Linux Test Project run all tests as well as
crash tests.
Athlon 1333 with 1024 MB RAM.
12 simultaneous "./configure && make -j 30" of the kde-2.2.2 source
packages (addons admin artwork base bindings games graphics libs
multimedia network pim sdk toys utils kdevelop.
This put a load of 40 - 90 on the system for the duration of the
run. (about 2 hours in total - only hits swap for a few minutes,
on this machine).
dbench 128 on Athlon
2.4.17rc2aa1 Throughput 41.7986 MB/sec (NB=52.2483 MB/sec 417.986 MBit/sec) 128 procs
2.4.17-rc2 Throughput 18.1516 MB/sec (NB=22.6895 MB/sec 181.516 MBit/sec) 128 procs
2.4.17-rc2 Throughput 20.4923 MB/sec (NB=25.6153 MB/sec 204.923 MBit/sec) 128 procs
2.4.17rc2aa2 Throughput 41.805 MB/sec (NB=52.2563 MB/sec 418.05 MBit/sec) 128 procs
2.4.17rc2aa2 Throughput 42.1931 MB/sec (NB=52.7413 MB/sec 421.931 MBit/sec) 128 procs
2.4.17-rc2rml1 Throughput 36.888 MB/sec (NB=46.11 MB/sec 368.88 MBit/sec) 128 procs
2.5.1-dj3 Throughput 26.363 MB/sec (NB=32.9537 MB/sec 263.63 MBit/sec) 128 procs
LTP regression
2.4.17rc2aa1 and 2.4.17rc2aa2 do consistently have a regression on one test, which
passes on 2.4.17-rc1 and 2.4.17-rc2:
nanosleep02 1 FAIL : Remaining sleep time doesn't match with the expected 4 time
nanosleep02 1 FAIL : child process exited abnormally
These aren't VM tests, per se. However, I found myself going back to this kernel
whenever I had a problem with 2.5.2-pre1.
All of the 2.4.17* kernels have been stable on my systems. For performance,
I give the nod to the aa kernels.
--
Randy Hron