On Mon, 2003-06-02 at 15:09, Greg Stein wrote:
> Actually, you probably *do* need to deal with Nagle. Apache disables the
> darned thing on its sockets. See the commentary in ap_sock_disable_nagle()
> in httpd-2.0/server/mpm_common.c.
I believe the Nagle algorithm only runs afoul of pipelined
implementations where data may be going in both directions at once.
Although I've referred to "pipelining" ra_svn recently, I may have been
using the term inaccurately; I've only been talking about switching the
direction of data transmission less often, resulting in fewer round-trip
delays. In all cases, either the client is writing and the server is
doing blocking reads, or the server is writing and the client is doing
blocking reads. The blocking reads should flush out any unsent partial
packets.
(If I'm wrong and we do need to disable the Nagle algorith, it's only
one line of code.)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Jun 2 21:33:42 2003