NetNews Usenet Archive 1992 #20/NN_1992_20.iso/spool/comp/dcom/cellrel/504

Text File | 1992-09-12 | 3.9 KB | 88 lines

Newsgroups: comp.dcom.cell-relay
Path: sparky!uunet!usc!cs.utexas.edu!sun-barr!ames!sgi!rhyolite!vjs
From: [e-mail] (Vernon Schryver)
Subject: Re: Another Application
Message-ID: <[e-mail]>
Organization: Silicon Graphics, Inc.  Mountain View, CA
References: <[e-mail]>
Date: Sun, 13 Sep 1992 06:01:42 GMT
Lines: 77

In article <[e-mail]>, bantha.decnet.lockheed.com!young writes:
> OK lets try this application;  A travel reservation service using the
> cell relay value added networks.
>
> In this case travel agents who want to set up a travel tour to, say,
> Central and Eastern Europe can send an exploratory wave to that region
> and visit a desktop computer in every hotel in the area, examine the
> hotel description (using some common data base application), returning
> any relevant info back to the travel agent setting up the tour.
>
> Here is how it works:
>
> The service uses a single VCI leased from the local value added net,
> valid over that whole region, and which spans all the hotels which are
> likely targets of for travel tours. Now at each hotel exists a software
> package which runs on some low cost, cheap, clone-able PC, and provides a
> common software format for hotel owners to describe the facility and local
> amenities.
>
> The service user prepares an exploratory wave, sends it (via TCP/IP) to
> the local wave launching point, lets it loose into the hotel spanning net.
> The wave collects data back to the launching point, where it is transferred
> back to the agent.
>
> Some few minutes later come back a series of digital brochures from a
> desktop in each hotel in the region.
>
> No hassle, no muss, no management, no extra layers.
> Matt Young
> Lockheed
> [phone]


Check out the literature for "reliable multicast."  And also look for
"scaling."

The problem of reliable datagram multicast is technically hard, except
in toy environments.  As far as I know, no one fiddling with frame
relay stuff is even remotely talking about such stuff.

If you know of a way to ensure that
    (1) absolutely every hotel hears the "wave"
    (2) the travel agent receives absolutely all of the answers
    (2) with "No hassle, no muss, no management, no extra layers",
then please write it up, or better, build a prototype implementation.
You'll be famous.

I know of some experiments and successes in reliable multicasting.
However, at least some are like XTP reliable multicast, and intended
for where it is good enough that most of the time most stations hear
everything, as in a military network in combat.

You have described a situation where any systematic failure of the wave
to reach a single hotel or for the answer from that hotel will invoke
the Dreaded Lawyers.  Lawyers are far less forgiving than enemy
solders.  Soldiers generally stop shooting when you're dead.

You have also described a situation dangerously likely to collapse any
network.  Consider the similarity between an innocenty formed,
insufficently specific query for "a bed in a hotel in North America on
July 4th" with what has come to be called "sendsys bombing."  That is
when someone sends out a USENET "sendsys" message requesting that all
receiving stations answer with their USENET connections.  Today that
happens about twice a week, when either someone soon to be less naive
does it to themselves, or a malfactor forges it for for someone else.
Back in ancient days when there were only a few thousand USENET hosts,
"sendsys" was good for building maps.  Today, it's good for receiving
hundreds of thousands of email messages in 10-40 hours.

I bet the number hotels and motels currently connected to on-line
credit card verifcation networks, and so likely to be customers of your
service is more than one order of magnitude higher than the current
number of USENET hosts.  You have described a system that might have
scaling problems.


Vernon Schryver,  [e-mail]