Ok, I've given up after a week and restarted squid using ufs
today at 13:30 in France.

Now, MRTG graphs are showing a lower CPU usage for our squid proxy,
as cachemgr.cgi (CPU Usage, 60 minute avg: 23.25%) and vmstat:

> vmstat 1 10
Virtual Memory Statistics: (pagesize = 8192)
  procs      memory        pages                            intr        cpu
  r   w   u  act free wire fault  cow zero react  pin pout  in   sy  cs us sy id
  2 111  32  48K 1509  12K  359K  74K 145K   278  44K    0  4K 187K  4K 21 40 39
  3 110  32  48K 1492  12K     5   16   18     0   28    0  3K  2K  3K 10 41 49
  3 110  32  48K 1458  12K     0    0    0     0    0    0  5K  4K  4K 19 25 56
  2 110  33  48K 1422  12K     0    0    0     0    0    0  4K  2K  4K 13 44 43
  2 110  33  48K 1403  12K     1    0    0     0    0    0  6K  4K  5K 23 28 48
  3 110  32  48K 1365  12K     0    0    1     0    0    0  5K  3K  4K 13 47 40
  2 111  32  48K 1367  12K     0    0    0     0    0    0  6K  4K  6K 20 30 50
  5 109  31  48K 1340  12K     0    0    0     0    0    0  6K  2K  5K 12 44 44
  2 111  32  48K 1326  12K     0    0    0     0    0    0  4K  3K  5K 13 22 65
  5 109  31  48K 1292  12K     0    0    0     0    0    0  6K  3K  5K 14 39 47

Curiously, top and dxsysinfo (a Dec tool) keep saying CPU load is
very high:
load averages:  1.39,  1.46,  1.38                                     14:55:52
62 processes:  1 running, 1 waiting, 14 sleeping, 46 idle
CPU states: 23.7% user,  0.0% nice, 65.4% system, 10.7% idle
Memory: Real: 297M/489M act/tot  Virtual: 1500M use/tot  Free: 9200K

  PID USERNAME PRI NICE  SIZE   RES STATE   TIME    CPU COMMAND
 3557 root      42    0  154M  151M WAIT   17:25 20.20% squid
 3578 squid     42    0 1776K  196K sleep   0:20  0.40% unlinkd
 3645 root      44    0 8920K 6086K run     0:00  0.20% top
...

Who's right ?
Any advice ?

My squid is really slow.

> I've upgraded from squid-2.5.STABLE4 to squid-2.5.STABLE5 on a
> Tru64Unix 5.1 alpha XP1000.
> 
> First I've started 2.5.ST5 Monday at noon using ufs.
> Then I've restarted 2.5.ST5 Tuesday at noon using diskd.
> 
> Just see at:
> http://www.univ-st-etienne.fr/criter/stats/mrtg/cpu.janon.html
> 
> CPU load is, say, twice the load of squid before using diskd.
> Also, nightly CPU load is really higher than squid-2.5.ST4
> 
> It seems squid is using CPU, not diskd:
> janon/root> ps aux
> USER        PID %CPU %MEM   VSZ  RSS TTY      S    STARTED         TIME COMMAND
> squid      3931 77.2 34.5  181M 176M ??       R      Mar 09    10:20:10 (squid)
> root          0 17.3  5.3  643M  27M ??       R <    Mar 08    08:55:50 [kernel 
> idle]
> squid      3953  1.5  0.1 2.63M 480K ??       U      Mar 09    18:45.20 diskd 40
> 25344 4025345 4025346
> 
> >From cachemgr.cgi:
>         CPU Usage, 5 minute avg:        77.92%
>         CPU Usage, 60 minute avg:       76.15%
> 
> I thought diskd would help our squid cache under heavy load
> (2,000+ clients, more than 60 requests/second, very long answers)
> 
> Any idea ?
> Thanks in advance.


-- 
_________________________________________________________________
Magali BERNARD - CRITeR - Universit� Jean Monnet de Saint-Etienne
23 rue du Dr Paul Michelon - 42023 St-Etienne Cedex 2 - FRANCE
Tel: 04.77.48.50.62


Reply via email to