On Tue, Mar 19, 2002 at 12:45:18AM -0600, Bob Tanner wrote:
> Anyone read anything about distributed compilation of the kernel?
	no, but the thought did occur to me the other day, as a possible
excuse to buy a Rocketcalc box. (www.rocketcalc.com). ;)
	
> I'm talking about distributing the compile of the kernel across idle boxes on
> your network.
	it's possible that just building a beowulf cluster (which is what
you're looking for) and substituting 'pmake' for 'make' will do the job for
some problems.
	from what I hear, tho, the linux kernel has some ugly build
dependencies that the current kbuild _usually_ gets right; but Owen <forgot
his last name> gave a presentation at a linux conference in April last year
(forgot which one) where he talks about problems that exist in the current
make dependencies (this is not the same thing as ESR's CML2; which is just a
tool to create the makefiles).
	so trying to distribute the build process across multiple machines
may be a bit hairier than you might think. that said, it's probably a
worthwhile thing to look into. I don't see that there is much difference
between 'make -j 14' on one box; and having a make process that distributes
those make processes across several machines which NFS-mount the repository.
	you might want to look at Mosix; and ask them if anyone has
experimented with parallel make processes for the linux kernel.

> Something ala pvm. NOT like a cluster.
	pvm is what beowulf clustering uses (tho it could be that I'm
equivocating here). you're probably thinking of other types of clustering;
for redundancy or network load-balancing, which most certainly aren't
appropriate here.

Carl Soderstrom.
-- 
Network Engineer
Real-Time Enterprises
www.real-time.com