home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky alt.amateur-comp:664 comp.theory:2927 comp.misc:5002
- Path: sparky!uunet!dtix!mimsy!kainen
- From: kainen@cs.umd.edu (Paul Kainen)
- Newsgroups: alt.amateur-comp,comp.theory,comp.misc
- Subject: Re: Is Packet Switching Technology the real secret of the Internet?
- Summary: A few words about a very complicated subject
- Keywords: economy of scale, throughput, simulation, system design
- Message-ID: <63713@mimsy.umd.edu>
- Date: 26 Jan 93 05:13:00 GMT
- References: <1jl6ulINN5ov@usenet.INS.CWRU.Edu>
- Sender: news@mimsy.umd.edu
- Followup-To: alt.amateur-comp
- Distribution: inet
- Organization: U of Maryland, Dept. of Computer Science, Coll. Pk., MD 20742
- Lines: 33
-
- As I understand it, Kleinrock really did three things. He showed that
- by breaking up the messages into packets the network achieves "economy
- of scale" and so performs more efficiently. He also noted that the
- lengths of messages entering the various nodes could be treated as if
- they were independent random variables for the purposes of simulation,
- even though these messages are actually not independent. Finally,
- when the density of traffic on a link exceeds a certain threshold,
- which is the <capacity> of the link, then the actual throughput of the
- link (or similarly of the network) decreases. Kleinrock used a combo
- of formal analysis via queueing theory, heuristic arguments based on
- experience with actual traffic behavior and computer simulation.
-
- A major difficulty with the packet switching approach today is that
- the services which are competing for resources are extremely different
- in character (size and tolerance for delay or error). System design
- is also made difficult by possible tradeoffs between smart terminals
- and transmission bandwidth. For example, a digital video on demand
- network ought to be able to "look ahead" to see that, say, a satellite
- transponder will be unavailable for another few dozen milliseconds and
- so hold the frame in its buffer. Finally, conflict resolution among
- competing intelligent agents combined with the sheer rate of
- technological change have made the job of scaling up the old Arpanet
- highly non-trivial. Maybe someone else can chip in here (oops) with a
- description of "Asynchronous Transfer Mode" switching and its relation
- to broadband services. Design is especially difficult since the types
- of services which will be using the system are not yet known. Imagine
- trying to design a house and being told at the last minute, "Oh, yes.
- We need to accomodate some dolphins, too!"
-
- L. Kleinrock's work is spelled out in a lucid little book,
- Communication Nets, McGraw-Hill 1964 and Dover Press 1972. A more
- complete reference is Kleinrock's 2 vol Queueing Systems, Wiley, '75
- or also see T. Saaty's classic on q'g theory.
-