performance of new code
Katsuhiro Kondou
kondou at nec.co.jp
Fri Feb 2 10:02:43 UTC 2001
I think my patch works fine without any problem so far.
There may be some problem which I didn't find yet, however.
I'll commit it soon.
For those who are interested in, here is performance
comparison between current and patched one. Testing box
is Solaris2.6 on Netra t1(440Mhz/1GBmem/5*18GB built in
two D130s for cycbuff), and get fed fullfeed(~3MBytes/sec,
~20art/sec).
current code:
storage method cnfs:
innd: ME time 600021 idle 184009(413886) hishave 19096(9760) hisgrep 524(297) hiswrite 63329(9463) hissync 61(948) artclean 83397(9241) artwrite 106358(9241) artctrl 2736(516) artcncl 2737(519) perl 24(9241)
PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
9933 news 1 10 0 66M 35M run 19:25 39.01% innd
new code:
storage method cnfs:
innd: ME time 600047 idle 298002(297523) hishave 17522(7634) hiswrite 52299(7634) hissync 39(764) artclean 502(7635) artwrite 89013(7634) artctrl 6(123) perl 12(7634) nntpread 74022(297523) artparse 53491(300175) artlog 471(8042) datamove 1236(310234)
PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
9988 news 1 22 0 64M 35M sleep 9:02 18.71% innd
storage method trash:
innd: ME time 600001 idle 443185(258339) hishave 2013(9497) hiswrite 49281(9497) hissync 57(951) artclean 472(9498) artwrite 260(9497) artctrl 18(1674) perl 7(9497) nntpread 55753(258339) artparse 37783(260868) artlog 573(10050) datamove 889(264566)
--
Katsuhiro Kondou
More information about the inn-workers
mailing list