Danek Duvall wrote:
On Tue, Feb 10, 2009 at 01:52:17PM -0800, Brock Pytlik wrote:
What was on pipes?
Oh. I thought you were talking about
5724 seriously suboptimal bufsize in cpiofile.py
/me crawls back into his cave
Aaah, no no. In some of the early work on "search v1/indexing shouldn't
take up 66% of the install time," I tried changing the buffersize for
reading and writing the single large index file. From my notes (and my
recollection), going from the default buffersize up to 128k took my
install/uninstall timing test suite from 7.5 mins down to 5 mins. So if
we're reading and writing large files to/from disk, I think this might
be a fairly large win. For example, reading writing the catalog.pkl
(5.7M) might be speedier. That was the only substantially large file we
create that I could find on the client machine in a few mins searching.
It might also be useful for reading and writing manifests. The largest
on my system was the openoffice-sdk one which topped out at 3.4M. I can
poke at that this afternoon and see if changing the manifest and catalog
pickle file buffersize gives any substantial time improvements.
Brock
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
