[Success (but still slow)] XEmacs 21.5-b16 "celeriac" (+CVS-20040301) i686-pc-linux
Stephen J. Turnbull
stephen at xemacs.org
Sun Mar 21 22:27:21 EST 2004
>>>>> "ms" == Michael Sperber <sperber at informatik.uni-tuebingen.de> writes:
Stephen> "Very" slow, no. But AFAIK the incremental changes are
Stephen> only supposed to keep the API the same.
ms> No, they were also tuned to keep the performance roughly the
ms> same. And our benchmarks have indicated that they do.
Say what you like about "benchmarks" (shall we revise Disraeli to the
more modern "lies, statistics, and benchmark results"? :-) both
Steve's personal experience and my (admittedly unreliable) profiling
results say 21.5 spends a _lot_ more time in GC, about 2-3 times for
me, and something like 20-30 times for Steve (but that may be due to
bad karma from bashing Windows, which are sorta like mirrors ;-). I
don't think any effort was made to adjust for other changes which
would affect user-visible performance, and that's what this thread is
about. I would consider worrying about it premature, except that
Steve's results are horrifying, we can't ignore them (can we?).
My belief is that 21.5 conses more, it also counts large strings as
consing which 21.4 does not (Martin Buchholz found that bug), so even
with GC having equivalent "raw" performance, 21.5 probably inherently
needs a faster GC, or a tuned-down collection frequency, to keep up
with 21.4 from the user's point of view.
Stephen> And there are no plans for a single new GC,
ms> Urmh---but there is. Marcus Crestani has already written the
ms> allocator, and is currently writing up his term project paper.
I was referring to Hrvoje's use of the word "the". I was vaguely
aware that one or more of your students was already writing a new
collector, but the Boehm collector is always of interest to somebody,
and Stefan Monnier has written a new collector for GNU Emacs. There
are several candidates, and unless Marcus hurries up, he won't
necessarily be first.
Institute of Policy and Planning Sciences http://turnbull.sk.tsukuba.ac.jp
University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Ask not how you can "do" free software business;
ask what your business can "do for" free software.
More information about the XEmacs-Beta