Sergey 'DoubleF' Zaharchenko wrote:

On Wed, 03 Mar 2004 00:34:37 +0800
"Zhang Weiwu" <[EMAIL PROTECTED]> probably wrote:

Today I did't experiment on {rt|id}prio(1), just to be interesting. I am trying to understand FreeBSD priority mechenism. IMHO realtime/idle priority only starvs processes when there is no cpu resource at all.

To test, I run mpg321 on realtime priority 3, run ppp on realtime
priority
4, a little lower than mpg321, and scp over the ppp tunnel on normal priority.

Top(1) shows an average about:
55.2% user, 0.0% nice, 21.5% system, 12.2% interrupt, 11.1% idle
* mpg321 takes average about 70%;
* ppp is seriously slowed down, it takes about 1.5% (normally 17% on full

speed);
* sshd almost starves; it takes 0.00% (normally 2%) and only transfer several bytes once several minutes; the otherside scp prompts "stalled".


sshd doesn't starve CPU, I guess; as you reach it through ppp, and ppp
gets slow, sshd does, too.

I'm the kind of newbie think 11.1% idle cpu time to be free cpu resource,

because I don't have idle process. I wish to know why, as there are still

11.1% free cpu resource, ppp and sshd doesn't get this share?


The most frequent cause is that they don't need the CPU so much as
something else, like disk I/O, etc. Could you please post the whole
output of top (and a snapshot of systat -vmstat, for example)?

Here is my systat -vmstat
=============================================================================

2 users Load 1.44 0.91 0.42 Mar 5 20:00
Mem:KB REAL VIRTUAL VN PAGER SWAP PAGER
Tot Share Tot Share Free in out in out
Act 9004 2144 62560 4228 2232 count
All 22104 3040 2307048 7032 pages 4
Interrupts
Proc:r p d s w Csw Trp Sys Int Sof Flt 6 cow 352 total
1 33 624 130 96 465 99 37 11312 wire 100 0: clk
6392 act 1: atkb
6.0%Sys 6.3%Intr 33.9%User 0.0%Nice 53.9%Idl 3356 inact 86 5: sbc0
| | | | | | | | | 1452 cache 6: fdc0
===+++>>>>>>>>>>>>>>>>> 780 free 31 7: ppc0
daefr 128 8: rtc
amei Name-cache Dir-cache 29 prcfr 5 11: rl0
Calls hits % hits % 1 react 13: npx
82 72 88 pdwak 2 14: ata
13 zfod pdpgs
Disks ad0 9 ofod intrn
KB/t 14.15 67 %slo-z 5072 buf
tps 2 51 tfree 14 dirtybuf
=============================================================================


And my top:
=============================================================================

last pid: 3451; load averages: 1.01, 0.72, 0.33 up 0+18:29:36 19:59:22
35 processes: 1 running, 34 sleeping
CPU states: 39.3% user, 0.0% nice, 9.5% system, 20.6% interrupt, 30.6% idle
Mem: 6512K Active, 3596K Inact, 11M Wired, 1616K Cache, 5072K Buf, 356K Free
Swap: 128M Total, 13M Used, 115M Free, 10% Inuse
PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND
3445 music -8 -104 5556K 884K pcmwr 1:43 36.08% 36.08% mpg123
2075 root 98 0 3856K 1292K select 7:43 6.79% 6.79% ppp
3448 zhangweiwu 96 0 6436K 2072K select 0:04 1.61% 1.61% sshd
3451 zhangweiwu 96 0 2212K 1220K RUN 0:02 0.93% 0.93% top
211 root 96 0 1812K 228K select 0:23 0.00% 0.00% dhclient
420 root 96 0 3488K 328K select 0:09 0.00% 0.00% sshd
271 root 96 0 1312K 228K select 0:03 0.00% 0.00% syslogd
444 root 8 0 1336K 188K nanslp 0:02 0.00% 0.00% cron
3426 music 5 -91 9936K 2704K ttyin 0:01 0.00% 0.00% cmp3
3450 zhangweiwu 4 0 2576K 1124K sbwait 0:01 0.00% 0.00% scp
3446 root 4 0 6220K 1688K sbwait 0:01 0.00% 0.00% sshd
373 root 96 0 1236K 84K select 0:01 0.00% 0.00% usbd
3415 root 4 0 6244K 0K sbwait 0:00 0.00% 0.00% <sshd>
3420 root 4 0 6244K 768K sbwait 0:00 0.00% 0.00% sshd
=============================================================================



Before this vmstat and top snapshoot, I tried ppp+ssh send a file, speed is average 40KB/s, cpu is mostly free. And I start the mpg123 at rtprio 20, ssh begin to speed down. 10 seconds since the music begin, ssh display higovdemo.avi 10% 2064KB 4.8KB/s - stalled -


What I don't understand is there are 30.6% idle cpu resource, enough bandwith, what starved ssh?

FWIW, splay uses <30% of a P150:)


_________________________________________________________________
免费下载 MSN Explorer: http://explorer.msn.com/lccn


_______________________________________________
[EMAIL PROTECTED] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "[EMAIL PROTECTED]"

Reply via email to