On 1/29/07, Roland Mainz <[EMAIL PROTECTED]> wrote:
Eric Enright wrote:
> On 1/29/07, Bruno Jargot <[EMAIL PROTECTED]> wrote:
> > On 1/29/07, Michelle Olson <[EMAIL PROTECTED]> wrote:
> > > Hi Rich,
[snip]
> Waiting a little extra time to receive the archive is far better than
> waiting a lot of extra time to unpack it.
Offtopic: Both gzip and bzip2 run much faster on SPARC, either via using
large pages (e.g. 64k or 4M; 512k pages will not work on Niagara1
machines... ;-( ) for the source and destination buffers and/or using
profile feedback compilation (and/or, XIPO (=interprocedural
optimizer)).
In theory it may even be "nice" to provide gzip and bzip2 as optimized
libraries to avoid that less-optimal versions are used by the various
applications (it would even allow to use some of the gzip-accerlator
cards out there (by replacing the libraries with versions which use this
hardware)).
Just for fun, I did a very quick-n-dirty benchmark on a 333MHz U10. I
tried 64k, and "default", and 64k is the winner. I also tried 4M but
results and possible long-lost memory make me wonder if 4M is
available on USIIi, so I omit the results (they were very similar to
"default".) Each test was done with the source file read over NFS and
written to /dev/null, and each test was done twice. The file was an
archive of /usr/X11R6 from FreeBSD, 145M originally and 66M
compressed.
cc -xO2 -xpagesize=64k
real 1m53.009s
user 1m46.689s
sys 0m1.984s
real 1m50.108s
user 1m46.299s
sys 0m1.507s
cc -xO2 -xpagesize=default
real 2m11.240s
user 2m5.738s
sys 0m1.775s
real 2m10.303s
user 2m5.548s
sys 0m1.663s
--
Eric Enright
_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org
