Re: GNOME performance [Was: the same page]

[jg pa dec com (Jim Gettys)]

I'd like to second, third, and fourth Alan's suggestions...

I'll add one tool to the list Alan provided: we have tools around for looking 
at X protocol traces. Sometimes people do very stupid things causing X 
to take unnecessary time....

I learned more from profiling X early on than everything else put together.
I expect that we could do likewise, with similar benefits, with
Gnome.  Gprof is an amazing tool when used well, and when you scratch
your head over both CPU time and procedure call counts.

We found outright bugs, stupidities, and subtle consequences of design 
decisions by the initial performance war Bob Scheifler and I had with 
Phil Karlton in the fall of 1984 than everything that came after.

The second round of performance wars came between Keith Packard and Joel 
McCormack on the X11 server in 1989.  By the end of those performance 
wars, X11R4 used 40% of the memory (excluding pixmap and backing store 
usage), than X11R2, while being amazingly faster.

There is alot to be said to making things truly "fast".

We have machines that are 100x faster than we did in the 1980's
(for CPU).  I think we'll find disk and memory touching the culprits
now.  Making things small is always worth it: things almost always
end up faster/better.
				- Jim


--
Jim Gettys
Cambridge Research Laboratory
Compaq Computer Corporation
jg pa dec com