On Fri, 2011-11-18 at 08:16 +0000, Tom wrote:
> Here's the stats from my cluster at the moment (8am) (these figures wll
> ramp up considerably!) (apologies if the html doesn't end up
> translating well!)
> 
> Server
> Load Avg
> Processed/Min
> Busy Child Proc
> Proc Time
> 10.44.219.192
> 0.34
> 42
> 1
> 0.31
> 10.44.219.193
> 0.19
> 42
> 1
> 0.32
> [...]
> 
> 
> 
> 
> 10.44.219.208
> 0.55
> 164
> 1
> 0.16
> Total:
> 0.45
> 1,527
> 16
> 0.21
> 
> and 15 minutes later :)
> 
> Server
> Load Avg
> Processed/Min
> Busy Child Proc
> Proc Time
> 10.44.219.192
> 0.40
> 63
> 1
> 0.34
> 10.44.219.193
> 0.59
> 61
> 1
> 0.38
> [...]
> 
> 
> 
> 
> 10.44.219.208
> 0.84
> 224
> 1
> 0.15
> Total:
> 0.66
> 2,174
> 16
> 0.23
> 
> 
> output from top, after running spamassassin for a couple of minutes:
> 
> [root@spam_209 ~]# top
> 
> top - 08:01:24 up 23:27,  3 users,  load average: 31.22, 10.73, 3.84

These numbers don't match up. The above, "after 15 minutes" claims load
average < 1, whereas this "after a couple minutes" skyrockets.

Also, I guess the "spam_209" hostname corresponds with the IP, so this
machine is not included in the above stats.


> Tasks: 115 total,  40 running,  75 sleeping,   0 stopped,   0 zombie
> Cpu(s): 99.0%us,  0.7%sy,  0.0%ni,  0.0%id,  0.0%wa,  0.0%hi,  0.3%si,  0.0%st
> Mem:   2055676k total,  1985400k used,    70276k free,    27704k buffers
> Swap:  4128760k total,      344k used,  4128416k free,   470672k cached
> 
>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
> 18458 spamd     20   0  255m  58m 3136 R  2.6  2.9   0:02.90 spamd
> 18463 spamd     20   0  255m  58m 3152 R  2.6  2.9   0:04.62 spamd
[...]

Here we have 40 spamd processes, each getting a tiny share of the CPU
resources.

I might be wrong, but since your -L local only spamd processes are
entirely CPU (and memory) bound, wouldn't it be more appropriate to run
about twice as much processes than CPU cores, to avoid context switches?

As it is right now, there are 5 (or even 10?) spamd processes per CPU
core, all busy burning CPU simultaneously.

Raising the number of concurrent spamd processes is useful to avoid idle
waiting for DNS, etc -- which you don't have.


-- 
char *t="\10pse\0r\0dtu\0.@ghno\x4e\xc8\x79\xf4\xab\x51\x8a\x10\xf4\xf4\xc4";
main(){ char h,m=h=*t++,*x=t+2*h,c,i,l=*x,s=0; for (i=0;i<l;i++){ i%8? c<<=1:
(c=*++x); c&128 && (s+=h); if (!(h>>=1)||!t[s+h]){ putchar(t[s]);h=m;s=0; }}}

Reply via email to