Re: Circuits? Circuits?

1996-03-28

Richard Moore

Someone asked:
>Circuits? Could someone please enlighten me on this?

        Each Internet message (packet?), if I understand it right, is
independently routed through the network.  The advantage is that network
management is trivial, a node can go down, and traffic automatically routes
around it immediately, without any central authority notifying everyone to
"Stop using node X".  This scheme works well for asynchronous applications
and has the advantage of simplicity and robustness.  (How does telnet fit
into this?)

        At the time Arpanet was being built, there was a competitor in the
lab at Tymshare, called Tymnet.  It was oriented around _sessions_ rather
than _packets_, and it built a "virutal circuit" whenever a user logged in
to whatever host he or she was accessing.   The circuit consisted of
coordinated table entries, stored in a sequence of computers that connected
the user with the host.

        The advantage of circuits is that data can be routed with very
little overhead as it dribbles (or surges) along the virtual circuit,
giving better response times to interactions, and better support for
continuous-trasmission of high bandwidth data.  The disadvantage is that a
higher-order of network self-awareness is needed to set up the virtual
circuits, and this carries the risk of central-failure meltdowns,  although
I imagine modern schemes have found ways to build in distributed
robustness.  Certainly, a telco won't tolerate its internal traffic being
disturbed by the outages suffered by an interconnected competitor.

        It is "circuits" and "isochronous protocols" that are dominant in
the design-center for the info-bahn.  Icochronous, by the way, means that a
given bandwidth is guaranteed, at a guaranteed latency, for a time period.
Thus if you're watching an action film, and don't want any breaks in the
motion, then you want an icochronous circuit (from the video-server to you)
rated at enough kilobaud to keep up with the compressed stream, and you
want the service guaranteed for the length of the film.

        When audiences are measured in the millions, even if many choose to
watch "broadcasts", providing reliable circuits on demand across
independent carriers is no mean feat.

        In the economics of such a system, where optimization of resource
utilization is critical to mission success, the underlying costing bases
would most likely include:
        - circuit-setup overhead
        - bandwidth and latency demanded
        - time of day
        - length of session
        - some distance-related factor

        In order to talk about how email would/should be priced, we need a
model of how internet-like traffic would be embedded in such a system, how
it would utilize circuits, etc.  It makes a difference whether we're
talking about Eudora-usage or remote-TTY-usage, btw.


Hope that helps,
Richard