Dan Nelson (dnelson_at_allantgroup.com) wrote: >In the last episode (Apr 14), Holger Kipp said: > >> currently, we have 11.3 web pages per second with SCHED_4BSD >> 11.5 web pages per second with SCHED_ULE >> with 0% idle under SCHED_4BSD, 5-7% idle under SCHED_ULE. >> (this is without witness or invariants, of course) >> (under 4.9-STABLE numbers are slightly below 11) > >What's the %User and %System? No amount of kernel improvements are >going to help you if your webserver does (1/11.5) sec of user CPU work >per page. Rest assured that we still have enough going on in %System to gain something from kernel improvements. Furthermore, sometimes lots of processes are in state 'GIANT', and top list many processes as 'locked'. Within the last week the number of web pages generated per second went up from 11.5 (SCHED_ULE) to 11.99 (SCHED_ULE). CURRENT of today gives the following (top, random snapshots): 2,76 0,73 75 5 54 16 67,2 26,4 1,0 5,4 15,44 5,86 75 8 54 13 69,3 24,5 1,0 5,2 17,35 8,28 74 12 54 8 68,6 25,9 1,0 4,6 16,60 9,78 75 21 54 - 68,9 25,8 1,2 4,2 16,70 10,66 75 3 53 19 68,5 26,0 1,4 4,2 15,75 11,75 75 5 52 18 67,9 25,1 1,7 5,4 load1 load2 proc running sleeping lock %user %system %irq %idle Usually many processes have state *Giant. As this is a dual XEON with Hyerthreading, I'm not sure if the %idle really gives correct values. I would expect some performance increase as soon as Giant is gone such that idle should be next to zero. Unfortunately I can't give you the latest performance values for CURRENT as of today, because system hangs right now (after running slightly more than 5000 requests according to ab). Keyboard is dead, ping gives no response. no output whatsoever. Nothing in /var/log/messages.Received on Thu Apr 15 2004 - 02:37:59 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:51 UTC