Dnia czwartek, 28 września 2006 09:24, Petter Haggholm napisał:

> contrast to all the recommended settings I have seen so far -- standard
> procedure seems to be -j$(no. cores+1), I presume so that one make

Ah, yes - one more thing. The estimation of -j for multi-core i've seen in 
various manuals varies from mentioned here #cores+1 up to 2*#cores. IMHO it 
depends on the very application being compiled.

* for large amount of small sources the I/O time is comparable to CPU time. In
  this case higher -j values are reasonable, as they overlay IO of one file
  with compilation of the another. As size of the files decreases, the -j may
  be increased

* for small amount of big sources the CPU time takes significantly longer than 
  IO time (and memory constraints become more important), so high -j values
  just cause time expensive process switching between concurent long-term 
  compilations (and might eat all the memory as in my case)


-- 
 Pawel Kraszewski
 www.kraszewscy.net

-- 
[email protected] mailing list

Reply via email to