Latest cvs-fast-export benchmark

[esr%thyrsus.com@localhost (Eric S. Raymond)]

This is the profiling sumary from cvs-fast-export's latest
run against netbsd src, on a Xeon E5-2660 v3 @ 2.60GHz:

       after parsing:	183.378          6170980KB
  after branch merge:	7015.202        18167564KB
               total:	12639.875       18255432KB
286747 commits/37062.508M text at 22 commits/sec.

That is 3 hours and 30 minutes, which is better than I was expecting.

Some of this is Laurence Hygate's memoized-sort patch; some of it is
me successfully reducing the working set, and some of it is building
with -march=native (this is exactly the kind of code for which that
is most effective).  It's almost exactly twice the throughput of my
first attempt.

And it's a factor of ten better than your numbers. Yeah, I guess we'd
better try to figure out why.

There's scope for reducing that time further.  Running over an SSD 
would, I estimate, knock it down to about two hours.  The "Great
Beast" I plan to build, based on a 3.5GHz Xeon ES-1650, might 
cut it to one hour and small change.

Build options for the Great Beast: http://esr.ibiblio.org/?page_id=6408

The Xeon-based proposal is the third one.  We have enough donations to
do it.
-- 
		<a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>