Andrew Pavlis wrote:
Ok, I understand that the kernal manages both the process distribution.  But
let's say I am running 10 game servers (for instance) which do not support smp.
Does Linux-smp distribute the load over both processors?  Or does it assign 1
proc here 1 proc there, etc?  Is there a way for me to help Linux :0] ?

As far as top is concerned, I am just wondering if there is a program that splits
the processor % up.....the version I have reports 3 processes that are utilizing
the cpu the most (the rest are negligable or less than 2% of the processor).
Process1 = 49.4% Process2=33.4%  Process3=27.8%

Now, I am no math whiz, but if you add that up, you get 110.6%!  So, whats up
with that?  I take it for granted that top just sees both processors as having a
top of 100%, giving the smp system a 200% total.  But, I would like to see
utilization broken up amongst both processors, so process1=24.7%proc1 24.7%proc2
(guestimation) process2=16.7%proc1 16.7%proc2 (guestimation) process3=13.9%proc1
13.9%proc2 (guestimation)
so that I can get a general idea of how the non-smp software is being allocated
by the kernel.  Is there any utility out that can break down processes on both
processors, or will have to script my own......

Thanks again!!!
-Andy
 

See:

http://www.redhat.com/mirrors/LDP/HOWTO/SMP-HOWTO-7.html#ss7.3

Here you can find a version of xosview able to split the load among processors.
Moreover, in the same page there is a version of procps (top, ps ...) that show
you utilization broken up among processors  (this is a dump of top from my machine:
 

  9:57am  up 19:15,  2 users,  load average: 0.09, 0.08, 0.02
61 processes: 59 sleeping, 1 running, 1 zombie, 0 stopped
CPU states:  1.0% user,  0.4% system,  0.0% nice, 12.6% idle
CPU0 states:  0.0% user,  1.0% system,  0.0% nice, 98.0% idle
CPU1 states:  0.1% user,  2.0% system,  0.0% nice, 96.0% idle
Mem:  127812K av, 119760K used,   8052K free,  55696K shrd,   4288K buff
Swap: 409536K av,    272K used, 409264K free                 53496K cached

  PID USER     PRI  NI  SIZE  RSS SHARE LC STAT %CPU %MEM   TIME COMMAND
 1989 aldinuc   12   0  1152 1152   916  0 R     3.8  0.9   0:00 top
 1431 root       3   0 27860  27M  1648  1 S     2.8 21.7   1:00 X
    1 root       0   0   472  472   408  0 S     0.0  0.3   0:08 init
    2 root       0   0     0    0     0  1 SW    0.0  0.0   0:03 kflushd
    3 root       0   0     0    0     0  1 SW    0.0  0.0   0:01 kupdate
    4 root       0   0     0    0     0  0 SW    0.0  0.0   0:00 kpiod
    5 root       0   0     0    0     0  1 SW    0.0  0.0   0:00 kswapd
    6 root     -20 -20     0    0     0  0 SW<   0.0  0.0   0:00 mdrecoveryd
 ....)
 

In addition (from the same page) PSET is able to force processes on CPUs,
making you able to statically assign tasks on PEs. It needs kernel patches and
recompilation.
 

Marco
 

-- 
       Marco      ___       ___    mailto:[EMAIL PROTECTED]
     Aldinucci   ('v')     (`v`)   http://www.di.unipi.it/~aldinuc
   Ph.D student ((   ))   ((   ))  DCS - University of Pisa - Italy
  -------------/-"---"---/-"---"------------------------------------
 


Reply via email to