[distcc] Re: speedup with >2 hosts

[SCHMID Stefan]

Hi Harold,

I excluded the local host, and I got the following for COBRA:

-j10: 495
-j15: 490
-j20: 488

compared to 5 hosts including local host:

COBRA_7_6_2   -j1             1857.332
COBRA_7_6_2   -j2             1039.243
COBRA_7_6_2   -j3             763.843
COBRA_7_6_2   -j4             696.498
COBRA_7_6_2   -j5             609.371
COBRA_7_6_2   -j6             549.385
COBRA_7_6_2   -j7             539.807
COBRA_7_6_2   -j8             514.258
COBRA_7_6_2    -j9            494.551
COBRA_7_6_2   -j10            505.846
COBRA_7_6_2    -j20           462.276
COBRA_7_6_2   -j30           460.046
COBRA_7_6_2    j40           455.205 

So excluding local host does not help me with 10 CPU's.

I attached the 1min-uptime plots on the different machines (without local 
host): localhost goes up very high, the other hosts hardly exceed 1...

All hosts are the same and have 2 CPU.

So there are three things I don't understand:
- why is the uptime on local host so much bigger?
- i can't hardly find a good formula for which -j-option to use. it seems 
that up to j100 its almost "the more the better"...
- why do my execution times vary so much (also for different -j options), 
even though i run the same thing over and over again?

Cheers, Stefan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ALLCOBRAUPTIME2.eps
Type: application/postscript
Size: 125296 bytes
Desc: not available
Url : http://lists.samba.org/archive/distcc/attachments/20040406/ada9ef91/ALLCOBRAUPTIME2.eps