[distcc] First results from distcc integrated cache

[Dan Kegel]

We've been experimenting with adding a cache to
distccd.   We benchmarked building the Linux kernel; the
results show both the promise and the limitations of the approach.
Here are the preliminary results (and take these with a grain of salt;
we might have a few problems in our test setup yet, given the
poor showing of distcc ):

150 sec  test system without distcc.
110 sec. normal distcc, eight-server cluster   (26% speedup over no
distcc - bleh!)
114 sec. caching distcc on same cluster, first run
90 sec caching distcc on same cluster, second run (40% speedup over no distcc)
->

The 4% slowdown on first run might be due
to computing the SHA-1 hash of the preprocessed
source code before sending it out..  We do that
because we need to fingerprint the sources to
decide which server to send them to.  But choosing
a server doesn't need such a good hash, so maybe
we'll try using adler32 for that and see if it helps.

Anyway, hopefully we'll have real numbers soon.
- Dan

--
Why won't Johnny run Linux?  See http://kegel.com/linux/comfort