Re: ra_serf violating editor API restrictions

On 8/13/07, Greg Hudson <ghudson@mit.edu> wrote:
> One connection with no waiting for round trips should be just as good.

A few issues:

- Many real-world HTTP servers have a small limit on how many
'pipelined' requests they can handle before they close the connection
(httpd default is 100; svn.collab.net used to be 10; code.google.com
seems to hover around 8-10.) This is a major drag on the pipeline
efficiency - the cost of restarting the connection when it hits this
limit is a PITA.

- Some sites have load-balanced SVN front-ends (give you one guess
here); if we distribute the traffic across a few servers, we minimize
the traffic that each server has to handle. This is good.

- WebDAV often requires some synchronous requests - this is especially
prevalent with commit 0 not so much with update. Right now, commit is
handled over one connection - only update uses multiple conns; but
if/when folks get around to it, this is gonna be an issue. (My
thought is to fork WebDAV in a way that truly permits async/pipelined
commits.)

HTH. -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Aug 13 18:39:11 2007