httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "William A. Rowe, Jr." <wr...@rowe-clan.net>
Subject Re: mod_proxy_http 2.1.5-HEAD connection pooling
Date Tue, 07 Jun 2005 22:49:58 GMT
At 11:33 AM 6/7/2005, Gregory Steuck wrote:

>    Bill> Well, if you are playing in 2.1-dev, you better be using ajp
>    Bill> in your scenario!  ajp13 is a binary protocol.  It consumes
>    Bill> far fewer bytes, and far less cpu, composing the headers and
>    Bill> decomposing them in coyote on the Tomcat side.
>
>I was playing in 2.1-dev to see if I can use HTTP instead of AJP. If I
>am to use AJP I will stick with 2.0 for now.

LOL - actually the connection pooling has been designed from
the ground up for multiple protocols, but is only effective when
the back-ends can provide feedback, as ajp13 does, and http does
not yet.  Some have kicked around the idea of a proxy load balancing
aware http response, with some extra x-load: headers which would
provide enough for mod_proxy_http to be deterministic.  Right now
only mod_proxy_ajp is reasonably deterministic about load, IIUC.

>    Bill> The connection pooling and load balancing scenarios are
>    Bill> primarily designed for 1:many scenarios.  Since that doesn't
>    Bill> apply to your case, the distinction is insignificant.
>
>The main point I was trying to make was that connection pooling did not
>seem to work with mod_proxy_http in the configuration that I described.
>Should I file it as a bug?

As I mentioned above, the HTTP response headers really don't provide
enough feedback to do anything other than round-robin or failover
of load.  Feel free to file it as an enhancement request, of course.

However, if mod_proxy_ajp doesn't perform the way you want, we aren't
likely to make progress addressing http pooling.  So again, I would
encourage you to look at mod_proxy_ajp, which I understood performs
faster than mod_jk, mod_jk2 or mod_proxy_http.

Bill



Mime
View raw message