Re: keepalives and proxies: a request and a proposal

Luigi Rizzo writes:
> On Sat, 18 Nov 1995, Roy T. Fielding wrote:
> 
> > Tunnel is defined in the lastest HTTP/1.0 specification (it is one of
> > the many definitions I added to explain some of the characteristics
> > of HTTP communication that is often ignored by implementors).
> > Use of a tunnel is compliant with HTTP, but requires that the HTTP
> > semantics not be changed by the presence of a tunnel.
Hmm. The word tunnel mean for me IP tunneled trough IPX or vice versa.
(One transport protocol tunneled trough another.) Not better to name
application level bridge?
> I'll look at the definition of tunnels before commenting more on this.
> 
> > Suffice it to say that IP addresses exist at (or below) the transport
> > level, and HTTP exists at the application level.  Using IP numbers
> > (or even hostnames) to define application-level behavior is wrong
> > because it won't work when the transport layer changes.  Therefore,
> 
> In the original message, and in further postings, Lorenzo and I
> talked about "unique identifiers" of the node (client, proxy or
> server), not IP addresses. I agree that addresses exist at the
> transport level, but there must exist some other form of unique
> identifier that is understood at the application level. Just use
> that.
I propose using {transport-protocol,node-name or address} pairs as unique
identifiers.
I guess Novell later or sooner will define http over SPX as did with
smtp/spx and telnet/spx.
As I am informed, Novell operates an IPX network number registry, which
can be extended (on the technical side with some appropriate extensions
to NDS) to operate like internet address registries.
An other argument that Novell plans (or already does) world-wide scalability
tests of NDS.
And the HTTP draft speaks about TCP/IP implementations of HTTP, but not
excludes implementations over other transport protocols.
As I see, support for other protocols needs some change in URL syntax too.
(the "//" host [:port] part should be renamed to network part and extended to
support multiple transport protocols having multiple node naming schemes).
I don't know yet, how the IPv4 -> IPv6 transition will be made, but that may
need some distiction, because the first needs a name to IN A lookup while the
second needs name to IN AAAA lookup.

Andrew. (Endre Balint Nagy) <bne@bne.ind.eunet.hu>

Received on Sunday, 19 November 1995 12:07:58 UTC