[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: Link latency and ra_svn/ra_dav performance

From: Greg Hudson <ghudson_at_MIT.EDU>
Date: 2003-10-23 09:01:25 CEST

On Wed, 2003-10-22 at 23:08, kfogel@collab.net wrote:
> We haven't looked into pipelining & Neon much yet; the above is just
> talk until we know more.

As Greg said, the current implementation of Neon doesn't support
pipelining. I just looked at the code to verify this;
ne_begin_request() reads the response headers before returning control
to the caller, and the code is very much not designed to handle multiple
ne_begin_request()s before an ne_read_response_block().

I think if we can get a significant checkout performance benefit by
screwing caches, we should do so as a temporary measure. We have no
reports of anyone using HTTP caches as far as I know, and it makes no
sense to massively penalize all of our users for the sake of none of our
users. Redesigning Neon or moving to a different client library is not
a pre-1.0-sized task.

I would even go so far as to say that we should temporarily screw the
WebDAV protocol for commits and imports, and just send the edit over a
custom report, if someone is willing to code that up. (Are big imports
all that frequent? Not normally, but it's the first thing you do in
many uses of Subversion, and up-front slowness is sometimes worse than
frequent slowness.)

To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Oct 23 09:02:20 2003

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.