Old tradindexed implementation removed
Russ Allbery
rra at stanford.edu
Sun Oct 6 04:50:32 UTC 2002
Forrest J Cavalier <mibsoft at epix.net> writes:
> Russ wrote, in part:
>> As of last Thursday, I've finally retired my old INN 1.7.2-insync-1.1d
>> reader server and am now running a CURRENT snapshot as of 2002-09-06
>> with the new overview code in production, so I hope it's stable. :)
> Please do post the timer stats before and after.
I think I did that a while back... here it is:
Code region Time Pct Invoked Min(ms) Avg(ms) Max(ms)
overview write 01:37:51.460 6.8% 244059 15.044 24.058 55.547
overview write 01:46:16.920 7.4% 237849 17.734 26.811 41.979
(the first is the old, the second is the new). So it got a little bit
slower. For my purposes, it's not noticable, but I don't know if that
will also be true for everyone else.
I think the very small change is pretty impressive given that the old code
rarely wrote index entries to disk and the new code msyncs all over the
place.
Now that there's a full load on the server, it's gotten a bit slower:
overview write 02:38:18.656 11.0% 246679 14.195 38.506 59.892
which isn't surprising, since everything running on that server slowed
down. But that number isn't comparable to the previous number.
1.7.2 was using overchan, so I unfortunately can't directly compare it to
the new code. I'm pretty sure that the new code is slower to write than
1.7.2 was, since there are two files to write instead of one, but it's a
lot faster for reading.
But in any case, here's the full statistics from the old server followed
by the full statistics from the new server. Note again, though, that this
is apples and oranges since the old server was a faster system
(dual-processor Ultra 2 compared to a single-processor Ultra 1) with
slower disk (SPARCstorage Array compared to an A1000).
The new system is going to be upgraded probably next Thursday.
INND timer :
Code region Time Pct Invoked Each
idle 19:37:15.223 81.7% 8081102 8.741
article write 02:51:10.063 11.9% 229672 44.716
article control 00:04:57.698 0.3% 12217 24.368
article cancel 00:00:41.156 0.0% 1331 30.921
crosslink 00:00:00.470 0.0% 26058 0.018
site send 00:00:57.156 0.1% 679541 0.084
history write 00:01:17.480 0.1% 235129 0.330
history sync 00:00:08.057 0.0% 45975 0.175
history lookup 00:19:24.707 1.3% 243674 4.780
history grep 00:00:32.838 0.0% 8067 4.071
INND timer:
Code region Time Pct Invoked Min(ms) Avg(ms) Max(ms)
article cancel 00:08:08.283 0.6% 18854 3.141 25.898 72.304
article cleanup 00:01:28.374 0.1% 246705 0.266 0.358 0.599
article logging 00:01:37.142 0.1% 258242 0.282 0.376 0.712
article parse 00:08:53.655 0.6% 3099421 0.110 0.172 0.364
article write 02:32:44.808 10.6% 246679 18.130 37.153 66.101
artlog/artcncl 00:00:07.582 0.0% 16783 0.000 0.452 1.687
artlog/artparse 00:00:00.004 0.0% 5 0.000 0.800 2.000
data move 00:03:41.179 0.3% 3268628 0.020 0.068 0.456
hisgrep/artcncl 00:02:19.674 0.2% 11737 0.000 11.900 34.615
hishave/artcncl 00:02:28.547 0.2% 18854 0.000 7.879 14.844
history grep 00:00:00.000 0.0% 0 0.000 0.000 0.000
history lookup 00:41:54.330 2.9% 251076 0.118 10.014 14.089
history sync 00:00:00.076 0.0% 309 0.000 0.246 1.000
history write 00:39:49.587 2.8% 246679 2.133 9.687 13.673
hiswrite/artcnc 00:00:59.596 0.1% 7117 0.000 8.374 22.154
idle 16:34:30.028 69.0% 3130985 7.492 19.058 45.351
nntp read 00:11:10.944 0.8% 3086187 0.146 0.217 0.664
overview write 02:38:18.656 11.0% 246679 14.195 38.506 59.892
perl filter 00:00:00.897 0.0% 246699 0.000 0.004 0.031
python filter 00:00:00.000 0.0% 0 0.000 0.000 0.000
site send 00:01:04.834 0.1% 493734 0.074 0.131 0.335
TOTAL: 24:00:32 23:49:18.196 99.2% - - - -
--
Russ Allbery (rra at stanford.edu) <http://www.eyrie.org/~eagle/>
Please send questions to the list rather than mailing me directly.
<http://www.eyrie.org/~eagle/faqs/questions.html> explains why.
More information about the inn-workers
mailing list