Re: Seattle minutes

Peter Deutsch (peterd@bunyip.com)
Thu, 7 Apr 1994 13:08:56 -0500

Message-Id: <9404071808.AA13482@expresso.bunyip.com>
From: Peter Deutsch <peterd@bunyip.com>
Date: Thu, 7 Apr 1994 13:08:56 -0500
In-Reply-To: "Daniel W. Connolly"'s message as of Apr 7, 12:20
To: "Daniel W. Connolly" <connolly@hal.com>
Subject: Re: Seattle minutes

Hi,

[ Daniel W. Connolly wrote: ]

> In message <9404071511.AA13038@expresso.bunyip.com>, Peter Deutsch writes:
> >
> >
> >> I agree that a single protocol should suffice, though I would
> >> prefer something with less packet overhead than whois++.
> >> (i.e. UDP rather than TCP)
> >
> >It would probably be nice to have only a single protocol,
> >although I don't see it as an absolute necessity at this
> >point. Certainly a UDP-based solution would be a win,
>
> Check out prospero:
>
> ftp://prospero.isi.edu/pub/prospero/
>
> in particular:
>
> ftp://prospero.isi.edu/pub/prospero/../papers/prospero/prospero-bii.ps.Z

Actually, we're big fans of Prospero here at Bunyip
Central. It forms the heart of the transport layer of the
archie system and is thus the protocol used by the various
archie clients when making queries. It's also spoken
between the various archie front ends and the archie
search process (eg. our telnet frontend, as well as our
new gopher and WWW frontends). We're also using the
Prospero file system itself for the internal data
architecture in our follow-on products (more indexed data!
Coming soon to a server near you!)

Prospero allows a lot of nice things. We like being able
to attach arbitrary attributes to a link and will
eventually be using other features, such as the
time-to-live/callback features, for data maintenance. It's
a nice system and we like it a lot, but I'm not sure it's
what is needed right this moment for URN->URL experimentation.

For those who haven't looked at it, Prospero uses a
reliable datagram protocol for transport that Cliff Neuman
developed (I _think_ he calls it RDP. Cliff?). This runs
over UDP, but takes care of such things as detecting lost
packets, retransmission, etc. It can also break large
collection of data into smaller packets for transmission
(seen an archie response lately? It's all going through
RDP). All in all, it's a nice piece of work.

The Prospero gang have also been most responsive to
suggestions and a big help with our own extensions to the
Prospero system to accomodate archie. I think it's a
system engineered for growth and we love it.

On the other hand, the Prospero protocol itself (not the
reliable datagram stuff) was originally intended as a
distributed file system protocol so there's things in
there we just don't need right now (callback,
time-to-live, etc). Implementing Prospero takes more
effort than implementing WHOIS++, which matters during a
prototyping effort. Of course, you can do what virtually
every archie client author has done and just strip out the
needed code from the existing Prospero release , but
someone would have to do the work, document it and let
everyone know. There is an effort there.

On the other hand, WHOIS++ has its RFCs out and seems to
be relatively complete for its intended task (in fact, the
task was one of the motivators for its creation). Multiple
implementations are available and experimentation can
start tonight.

Passing WHOIS++ queries over Prospero's RDP may make sense
in the long run (although I'm not sure how the HOLD option
would be handled here). Somebody would have to look more
closely at the question than I have to decide if this made
sense. Meanwhile, I think we should be deploying some data
using existing tools.

- peterd

-- 
------------------------------------------------------------------------------
  My proposal for funding the Internet is pretty simple. I vote we institute
  an "Information Superhighway" tax, the proceeds of which will be used to
  fund network infrastructure. The way this would work is simple - every time
  someone uses the words "Information Superhighway" or any of its derivatives
  we strike them with a sharp object and make them pay a $10 fee (of course,
  the sharp object is not actually needed to make this scheme work, it's just
  in there because it seems an appropriate thing to do...)
------------------------------------------------------------------------------