home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.modems
- Path: sparky!uunet!wupost!gumby!destroyer!terminator!honey
- From: honey@citi.umich.edu (Peter Honeyman)
- Subject: Re: Modems, compression, and SLIP or PPP (was Re: Using Telebit PEP with SLIP)
- Message-ID: <1992Aug23.203750.22670@terminator.cc.umich.edu>
- Sender: news@terminator.cc.umich.edu (Usenet Owner)
- Reply-To: honey@citi.umich.edu
- Organization: Center for Information Technology Integration, Univ of Michigan
- References: <1992Aug23.202309.22443@terminator.cc.umich.edu>
- Date: Sun, 23 Aug 1992 20:23:09 GMT
- Lines: 59
-
- sorry about the earlier, abbreviated version of this article.
- i sure wish xrn would let me cancel ... oh well.
-
- Bob Sutterfield writes:
- |> Whether you use V.42bis depends upon your application. It will only
- |> very rarely (if at all) decrease your throughput, likewise latency, so
- |> it's probably safe to turn it on and leave it on for most purposes.
-
- in experiments run here, we have seen very high latencies due to v.42
- and v.42bis. (but since we use the network more for file transfer
- than interactive traffic, we do leave them on, as bob suggests.)
-
- |> A typical PPP frame is exactly three bytes larger than a typical SLIP
- |> frame. This expands a typical VJ-compressed telnet packet from five
- |> bytes to eight. So far as I can tell, the difference is unmeasurable
- |> and imperceptible.
-
- a typical vj-compressed telnet packet is four bytes, not five -- here
- are three quotes from rfc 1144, "compressing tcp/ip headers for
- low-speed serial links", by van jacobson:
-
- from section 1 (introduction): the average compressed header is 3 bytes
-
- from section 3.1 (the basic idea): Recognizing a couple of special
- cases special cases will get us three byte headers for the two most
- common cases---interactive typing traffic and bulk data transfer
-
- from section 3.2.2 (compressed packet format): compressed terminal
- traffic usually looks like (in hex): 0B c c d, where the 0B indicates
- case (1), c c is the two byte TCP checksum and d is the character
- typed.
-
- |> Few connections are error-free. A V.32bis/V.42/V.42bis modem, clocked
- |> at 38400 or better, can easily deliver bits fast enough to overwhelm
- |> primitive serial I/O interfaces such as are found in many popular UNIX
- |> systems. Those UART overruns are manifested as line errors, which
- |> PPP's 16-bit CRC (not a checksum!) will almost always catch, so that
- |> the PPP can drop the damaged packet before passing it up to IP. This
- |> will save your bacon if you're running non-UDP-checksummed NFS, even
- |> over a V.42 or MNP4 modem.
-
- there are many errors here.
-
- my home computer is an ibm rt, three or so mips, with a run of the
- mill 16550 async card. i run the interface at 57,600. uart overruns
- are infrequent -- way less than 1%. if my dinky, five-year old unix
- box can do it, so can yours.
-
- crc *is* a checksum -- that's what the second c stands for.
-
- i fail to see the point in having the link layer drop the packet.
- even if it gets past the ip header check, the transport layer will
- surely drop it. the ppp checksum is entirely redundant, wasteful.
-
- should i comment on the utter stupidity of running nfs (or *any
- networked application!*) over a non-checksummed udp? especially in a
- low-speed environment, where computational overhead is slight?
-
- peter
-