And I forgot one thing.
Wouldn't it be cool if the jobs are spawned in function of the system lod
average or CPU load average?
The servers could send the client the current load and so the next job could
be sent depending on that. Or the spawn could be done in pure round robin, or in pure
random. All I try to ask/propose is to be able to spend more than one job on each
machine.
And so we could spawn N processes on each machine depending only on the make
-jN, not the number of DISTCC_HOSTS defined (One DISTCC_HOST could well be a 4 CPU or
HyperThreading machine).
What do you think?
-jec
-----Original Message-----
From: Jean-Eric Cuendet
Sent: Tue 19/11/2002 10:45 AM
To: [EMAIL PROTECTED]
Cc:
Subject: [distcc] Multi CPU machines
Hi,
I just discovered distcc. It's a great tool, thanks.
I poked in the code and found that there is only one job spawned for
each DISTCC_HOSTS, right?
I found that because there is one lock file in DISTCC_TMP directory
for each host defined.
This prevent distcc from using all the CPUs available in a machine.
Is there a way to tell distcc to spawn more jobs by machine?
Spawning more than one job on one CPU is often faster than one job/CPU
(make -j4 is always better than make, even on monoprocessor machines...), so this is
legitimate.
Thanks.
-jec
vqjjrjjv
��\rf��)��+-v+-q�b��,jfڢ�!���X�������܂&����Y���b�ا~�݊�\
