[distcc] Contributing to distCC

Martin Pool mbp at sourcefrog.net
Mon Dec 13 22:37:24 GMT 2004


(One point: unix-oriented open source projects generally consider it
polite to post in plain text and not top-post.)

On 13 Dec 2004, Assaf Lavie <lavie at runbox.com> wrote:

> You mentioned Java. It sounds like a cool idea. Do you mean that we
> adapt the existing project to work with Java instead of GCC by, I
> guess, providing a gcc-like proxy for the Java compiler? Separating
> Java preprocessing (or the equivalent) from compilation?

I don't think a gcc-like proxy would work well.  Java doesn't have a
preprocessor; rather it reads interfaces from classfiles.  You
probably need a distributed shared view of all the classfiles.  

I don't know if the results would be much better than just using say
OpenMOSIX but it might be an interesting project.

> Generalizing distCC from "just" C compilation to general distributed
> batch processing appears to me, prima facie, not to conform with our
> requirements, since there are already numerous (open-source) grid
> computing projects out there.

Good point.

> About improving the scheduling - we have at our disposal a lab with
  more than a few computers. Did you have any specific ideas in mind
  for improving scheduling that might give us a first step in this
  direction?

Have a look in doc/ in the source tree.

> We had an idea, albeit somewhat 'academic' in nature, to further
  dissect the compilation process and see if that can improve the
  build speed. To dive into GCC, separate the different parts of the
  compilation process itself, and perform some of them in
  parallel. Just a rough idea at the moment, to be honest. But if you
  see any merit in it we'd be very glad to hear your opinion.

Cutting up gcc might be interesting, but is probably hard and will
take a lot of work to maintain as gcc changes.  I don't think the
payoff would be very good (but I might be wrong).  The nice thing
about distcc is that it does not require intrusive changes to gcc.

You could try writing a parallelizable linker.  That would be very
cool indeed, though potentially would take a lot of learning about how
the linker works.

-- 
Martin 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.samba.org/archive/distcc/attachments/20041214/452bb040/attachment.bin


More information about the distcc mailing list