Hi Folks:
I have a set of four nameservers. Each of these nameservers are
running identical kernel's except for a recent nameserver that is
running the current redhat errata kernel 2.4.20-20.
Anyways judging by the output of sar my 2.4.20-20 kernel is attrocious
compared to kernel-2.4.18-24. This puppy seems to be swapping
constantly in and out....
Linux 2.4.20-20.7 10/07/03
02:00:02 pgpgin/s pgpgout/s activepg inadtypg inaclnpg inatarpg
02:10:04 35.27 64.79 52533 2908 1212 11751
02:20:01 33.04 64.74 52646 3215 696 11756
02:30:00 24.47 54.18 51839 3709 720 11755
02:40:01 26.10 58.40 52369 2923 1220 11755
02:50:01 27.10 56.83 52488 2912 1199 11751
03:00:01 23.34 50.10 51819 3210 853 11746
03:10:04 28.64 53.98 52015 3140 1174 11755
03:20:01 28.87 65.76 52629 3014 919 11756
03:30:02 24.11 54.14 51815 3347 756 11759
03:40:01 27.58 70.18 50095 4130 1173 11753
03:50:01 26.17 62.57 50953 3808 1013 11756
04:00:00 24.20 57.09 49452 4932 695 11755
04:10:06 232.99 191.56 51957 2946 1164 11679
04:20:00 113.42 128.04 52745 2924 836 11742
04:30:02 40.21 71.77 51701 3422 812 11758
Average: 48.32 73.78 51804 3369 963 11748
Linux 2.4.18-24.7.x 10/07/03
02:00:00 pgpgin/s pgpgout/s activepg inadtypg inaclnpg inatarpg
02:10:00 1.07 10.49 48209 10138 1256 11920
02:20:00 0.07 10.23 47596 10492 1368 11891
02:30:01 0.19 10.42 48145 10870 848 11972
02:40:01 0.19 10.50 47441 11196 968 11921
02:50:00 0.87 10.76 47607 9516 2414 11907
03:00:00 0.10 10.56 47576 9885 2034 11899
03:10:00 1.11 10.94 48023 10227 1367 11923
03:20:00 0.11 10.28 47974 10533 1152 11931
03:30:01 0.15 9.60 47449 10807 1337 11918
03:40:00 2.32 10.69 47589 9330 2142 11812
03:50:00 0.14 10.25 48221 9840 1536 11919
04:00:01 0.10 10.46 47335 10347 1716 11879
04:10:02 1591.78 142.24 49940 1166 6683 11557
04:20:00 496.65 15.82 48275 1243 8319 11567
04:30:00 7.51 10.24 48663 2132 7633 11685
Average: 149.81 20.25 48003 8515 2718 11847
I'm running all servers on identical hardware. Redhat 7.3. I've used
hdparm to tune the harddrives (and they are similar harddrives), and can
get a similar score via hdparm -tT /dev/hda from both systems. Load
isn't substantially different. Workload that they do nightly is for
the most part the same. During nightly processing though, the poor
performing box stops responding to DNS queries while it backs up on it's
poor I/O performance.
Anyone have any suggestions. I'm going to go back to a previous kernel,
until I can get some leads on where else to look.
2.4.18-24.7 = vm.max-readahead = 127
2.4.20-20 = vm.max-readahead = 31
elvtune has different values, but even adjusting those values doesn't
seem to matter
2.4.18-24.7 = /dev/hda elevator ID 1
read_latency: 64
write_latency: 8192
max_bomb_segments: 0
2.4.20-20 = /dev/hda elevator ID 0
read_latency: 8192
write_latency: 16384
max_bomb_segments: 6
2.4.18-24
/dev/hda:
multcount = 16 (on)
I/O support = 0 (default 16-bit)
unmaskirq = 0 (off)
using_dma = 1 (on)
keepsettings = 0 (off)
nowerr = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 2491/255/63, sectors = 40031712, start = 0
busstate = 1 (on)
2.4.20-20
/dev/hda:
multcount = 16 (on)
I/O support = 0 (default 16-bit)
unmaskirq = 0 (off)
using_dma = 1 (on)
keepsettings = 0 (off)
nowerr = 0 (off)
readonly = 0 (off)
readahead = 8 (on)
geometry = 2491/255/63, sectors = 40031712, start = 0
busstate = 1 (on)
Drives are IDE
Thanks,
Brian Seppanen
seppanen@chartermi.net
