[distcc] distcc with xinetd... how?

[Martin Pool]

On  8 Aug 2002, Brian Hall <brianwhall at pcisys.net> wrote:
> Oh, I still have a question. I run tmpfs on all my systems, hoping for
> a small speed gain. I notice that it doesn't support sendfile, and
> distccd falls back to another method. I am thinking that there should
> be a slight speed hit because of this fallback, judging by the logfile.

I suspect that read/write from tmpfs will be roughly as fast as 
sendfile() from a regular partition, but I haven't measured it.
If you want to, please let me know what you find, but if not I wouldn't
lose any sleep.  Either way I doubt it will be the dominant factor.

> Now, should I point distccd to a real filesystem or what for the
> best distccd performance? Should there be a --no-sendfile option if it
> is known in advance that sendfile is not available?

I guess I should put one in to allow comparative measurements.  It's not a
big deal though because trying it and failing only takes a single system
call. 

> BTW, at the moment my client and servers on running on Mandrake 8.2,
> all i686 architecture (mix of Pentium Pro and PIIIs).

By the way, if you could give me some rough numbers on how much distcc
speeds up your compilation I'd appreciate it.

--
Martin