Re: "end-to-end" headers in HTTP/1.1 [was Re: Mandatory]

From: Henrik Frystyk Nielsen (
Date: Fri, Apr 17 1998

Message-Id: <>
Date: Sat, 18 Apr 1998 00:38:39 +0900
To: (Ted Hardie)
From: Henrik Frystyk Nielsen <>
Subject: Re: "end-to-end" headers in HTTP/1.1 [was Re: Mandatory]

At 17:30 04/15/98 -0700, Ted Hardie wrote:
>I think the key term in what you write here is "out of band"
>agreements; in the previous uses, the end user had some out-of-band
>way to know how the proxies in the chain would behave.  As is pretty
>obvious, that doesn't scale very well, and it doesn't allow anyone
>to change the behavior on the fly.  Mandatory gives us a chance
>to create an in-band method for providing that infomration.  The
>requirements are different for in-band methods, though, than out
>of band methods.  Some method of indicating that a Mandatory
>header is meant to be consumed by a specific proxy or class
>of proxies seems to me to be part of that.  Minimally, that
>would help deal with situations where a header meant to be
>consumed by a proxy isn't.

Any knowledge about proxies down the message chain except for the very
first proxy is already out-of-band - there is no way a client can get
knowledge about the third proxy without out-of-band communication or some
extension (which in this context is the same thing).

HTTP has two scopes: hop-by-hop and everything else (which may or may not
make sense to call end-to-end). Mandatory doesn't (and I believe shouldn't)
change this - it just allows applications to express how they extend the
message with either of these scopes.


- hop-by-hop which is between an adjacent HTTP client and server. Tunnels
do not count as a "hop" as they do not take any part in the communication

- everything else which we often call end-to-end. 
Henrik Frystyk Nielsen,
World Wide Web Consortium