W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > January to April 1999

Re: Proxies and persistent connections

From: Jim Gettys <jg@pa.dec.com>
Date: Fri, 23 Apr 1999 12:57:21 -0700
Message-Id: <9904231957.AA26628@pachyderm.pa.dec.com>
To: Wham Bang <wham_bang@yahoo.com>
Cc: http-wg@cuckoo.hpl.hp.com

> From: Wham Bang <wham_bang@yahoo.com>
> Resent-From: http-wg@hplb.hpl.hp.com
> Date: Fri, 23 Apr 1999 12:43:09 -0700 (PDT)
> To: http-wg@cuckoo.hpl.hp.com
> Subject: Proxies and persistent connections
> -----
> Hi all,
> 
> Me and a colleague are starting to work on upgrading a proxy server
> product from HTTP/1.0 to HTTP/1.1 and in the course of a design
> discussion, a question came up.
> 
> Upon the opening of a persistent connection from a user-agent to a
> proxy, is it OK for the user-agent to send multiple requests *for
> different hosts* down this same connection to the proxy?

Of course.

What you should not do as a proxy is try to simultaneously use the
same connection on behalf of two different clients; otherwise, you could
have denial of service attacks.

Serial reuse of a connection is just fine. (anotherwards, as soon as you
get all responses back from a set of requests, you are free to reuse that
connection on behalf of another or the same client.

> 
> My reading of the spec makes it seem that it is indeed OK, but I don't
> have an HTTP/1.1 proxy handy to be able to see how browsers behave.
> But it seems almost mandatory since clients are encouraged to open
> no-more than two connections to a given proxy...  Does that imply that
> proxies should/must be able to "multiplex" responses - ie send
> multiple HTTP/1.0 and HTTP/1.1 responses from various servers in the
> proper order into a single persistent connection with a client?

Yup.

Must be in order, as HTTP has no request number to allow a client to
get out of order responses.  This is a basic "bug" (some have claimed
feature, I, and many other claim bug) in HTTP design.

> 
> This of course increases proxy implementation complexity quite a bit
> wrt the other possibility, which is: one (or two) persistent
> connection(s) per origin server (even through proxies)...
> 
> Can anyone help clarify?
> 
> 
> BTW, I there is a typo in draft-ietf-http-v11-spec-rev-06.txt, section
> 8.1.4, last paragraph. "A single-user client SHOULD NOT maintain more
> than connections with any server or proxy." should, I think read "A
> single-user client SHOULD NOT maintain more than 2 connections with
> any server or proxy."
> 
Yup; we know about the typo; will be fixed in the RFC.
			- Jim
Received on Friday, 23 April 1999 20:59:53 EDT

This archive was generated by hypermail pre-2.1.9 : Wednesday, 24 September 2003 06:33:30 EDT