>
> Linux-2.1.15 is out, and contains most of the mega-patches. Can people
> confirm that SMP works ok again in the 2.1.x series?
>
> Linus
>
2.1.15 Benchmark Results
The kernel is stable, recompiles itself quickly, and the only gcc WALL warnings
I see are in drivers/sound/{soundcard.c,sound_switch.c,sb_common.c}
referencing unused or defined but not used variables.
Compiled with
gcc -D__KERNEL__ -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer \
-D__SMP__ -D__SMP_PROF__ -pipe -m486 \
-malign-loops=2 -malign-jumps=2 -malign-functions=2 -DCPU=686 -D__SMP__
I do not use -fno-strength-reduce as I have a patched gcc 2.7.2.
Disk I/O is AHA-2940 Ultra-WIDE (PCI-bus) to Seagate ST32155W without
AIC7XXX_TAGGED_QUEUEING or AIC7XXX_PAGE_ENABLE. I have used those in
the past and had success and some performance improvement, but my current
focus is on SMP in 2.1.
All the benchmarks are taken on a quiescent machine after reboot, running
XF86_S3, fvwm95-2, and an xterm.
# final med_file small_file context shell
#RAM CPU Version Date | execl | large_file pipe | process | syscall
# | | | | | | | | | |
128M-1CPU-LX2.1.15.961213 70.5 61.5 78.5 98.0 58.6 93.1 127.8 205.3 100.0 65.0
128M-2TLB-LX2.0.20.960917 48.1 54.3 60.6 63.1 53.2 46.2 36.9 95.9 141.7 25.6
128M-2CPU-LX2.1.15.961212 50.5 53.8 64.6 71.0 52.5 63.3 38.4 73.9 140.0 34.5
128M-2TLB-LX2.1.15.961212 52.1 56.3 64.9 69.8 51.7 63.7 38.4 95.1 148.3 34.4
The 2TLB is SMP with CLEVER_SMP_INVALIDATE #defined in include/asm/pgtable.h.
It really does work.
Everything seems normal, just faster than 2.0.
Now what was Alan saying about horribly clever FPU stuff? Context switch
seems to be the bottleneck.
Sam Chessman (SSC3) chessman@wauug.erols.com