home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!decwrl!sgi!rhyolite!vjs
- From: vjs@rhyolite.wpd.sgi.com (Vernon Schryver)
- Newsgroups: comp.dcom.cell-relay
- Subject: Re: [packet sizes]
- Message-ID: <nodr79s@rhyolite.wpd.sgi.com>
- Date: 25 Jul 92 19:15:26 GMT
- References: <2796@ucl-cs.uucp> <1992Jul24.195100.20475@nntp.uoregon.edu>
- Organization: Silicon Graphics, Inc. Mountain View, CA
- Lines: 45
-
- In article <1992Jul24.195100.20475@nntp.uoregon.edu>, jqj@duff.uoregon.edu (JQ Johnson) writes:
- > It's not clear to me that VJ compression is really appropriate for ATM
- > nets. VJ compression, as I understand it, gets most of its power by
- > observing that the path over serial circuits is typically used (mostly)
- > for a single TCP bidirectional stream at a time. Hence you can predict the
- > next packet, and encode the whole header essentially as a stream
- > identifier plus a single bit that says "the next packet is in fact the one
- > you are expecting", with only a small percentage of your packets missing
- > the cache of current streams and having to be encoded long-form. VJ
- > compression doesn't seem to me to do as much good if the physical channel
- > is muxing lots of independent packet streams. For example, it's not of
- > much use, I don't believe, on a high speed path between two routers busy
- > with lots of low-speed connections.
- >
- > I'd be interested in the comments of other people more familiar with
- > TCP/IP compression schemes than I am. How good IS the match between
- > VJ compression and ATM?
-
-
- No, VJ compression works fine for several simultaneous virtual
- circuits. Any real TCP network often has more than one simultaneously
- active circuit even over a SLIP line. That's why anyone sitting down
- to invent something like VJ compression without looking at the standard
- or his code does just as he did, and includes some kind of circuit or
- session or context ID in the compressed header.
-
- You can have as many simultaneous VJ-like-compressed circuits as you
- wish to devote resources at the ends and bits in the compressed
- header.
-
- "VJ-compression" is not really about "compression." It is more about
- "header prediction", with the near side sending only those bits that
- the far side might not be able to predict.
-
- I wonder about doing VJ-compression at full speed over ATM. Some XTP
- ideas included finding the right "context" as the head of the packet is
- received, and using that "context" to forward, ACK, etc the packet,
- sometimes before you see the trailer. That does not seem trivial at
- gigabit rates on a stream of tiny packets without some special
- silicon. Last I knew, PEI was doing well with continuous streams of
- minimal sized UDP and TCP packets at FDDI rates. I do not know how
- they are doing at high speed.
-
-
- Vernon Schryver, vjs@sgi.com
-