Re: persistent connections/pipelining (was Re: Moo Meeting transcript)

Hi,

Kolics Bertold, University of Veszprem wrote:
> - I am a bit confused about persistent connections since the spec (RFC
> 2068, section 8.1.3) says:
> "A proxy server MUST NOT establish a persistent connection
> with an HTTP/1.0 client.

> However, 99% of the clients claims itself to be HTTP/1.0 compliant.

They're not daring enough to announce that they're 1.1. :)
Actually I think it's because the majority of web servers will not understand
the HTTP/1.0 request. Apache is one of the few.

> But in another section of the spec (19.7.1) there is a sentence: "An
> HTTP/1.1 server may also establish persistent connections with HTTP/1.0
> clients upon receipt of a Keep-Alive connection token."

In HTTP/1.0, Connection: Keep-Alive is used to specify that the client wants
a persistent connection. HTTP/1.1 on the other hand persists by default,
until you give a Connection: close header.

> - there is a problem with Netscape 2.x (and compatible) browsers with
> keepalive settings (see Apache_1.2b7). These browsers should not be served
> with persistent connections.

We can introduce something like:

acl mozilla browser Mozilla
persistent_access deny mozilla
persistent_access allow all
 
> - I suggest using an extra configuration option: timeout value for
> persistent connections.

Definitely.

> - the easier part is implementing the client-squid, squid-squid
> connections. Squid-server is a bit more difficult since the content
> negotiation part of HTTP/1.1 must be implemented as well.

I think that Squid to Squid should be done first, since we have control
of both sides of the code. I think this would also lead to the greatest
benefits in
 
> - there was a suggestion for turning Pragma: no-cache to IMS request in
> Squid. The same question should be raised for HTTP/1.1: should clients'
> Cache-control: max-age=0 to be translated to an IMS request?

No, I think not. no-cache is more of a hack, it seems, especially since
the Pragma: header has been deprecated. max-age is more of a real request.
Do any browsers actually use max-age?
 
> There are also issues in HTTP/1.1 according to caches (see
> <URL:http://www.w3.org/pub/WWW/Protocols/HTTP/Issues/>)

I am loading this now...

Regards,

-- 
miguel a.l. paraz  <map@iphil.net>                              +63-2-893-0850
iphil communications, makati city, philippines          <http://www.iphil.net>