Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!aplcen!uunet!mcsun!ukc!strath-cs!baird!jim From: jim@cs.strath.ac.uk (Jim Reid) Newsgroups: uk.misc,eunet.followup,comp.protocols.tcp-ip Subject: Re: Internet routing Europe - USA -} Europe... Message-ID: Date: 7 Sep 90 12:26:57 GMT References: <1990Aug30.091435.1982@ircam.ircam.fr <6190@vanuata.cs.glasgow.ac.uk> <4847@tardis.cs.ed.ac.uk> Sender: jim@cs.strath.ac.uk Organization: Computer Science Dept., Strathclyde Univ., Glasgow, Scotland. Lines: 64 In-reply-to: anarchy@tardis.cs.ed.ac.uk's message of 6 Sep 90 19:04:35 GMT In article <4847@tardis.cs.ed.ac.uk> anarchy@tardis.cs.ed.ac.uk (Alan Cox) writes: Firstly remember that when Janet was first setup properly they were dealing with a wacky array of bizzare processors, and also that then tcp/ip was abit fof an experiment. Unfortunately they haven't woken up since. Also they have a reasonable argument at the moment about relative efficiency of coloured book and tcp/ip. Discussions on the relative efficiency of network protocols tend to have an air of 'how many angels can dance on the head of a pin' unreality. The Coloured Books may be "more efficient" over JANET when compared with Internet protocols, but how can anyone sensibly compare the two? It's like comparing apples and oranges. First consider the network architectures. CB works on top of X.25 which more or less guarantees a reliable end to end connection. Internet protocols were developed for networks which may drop packets or deliver them out of sequence. [Let's also note that CB only offers connection-oriented services. The Internet world can also provide datagram and multicast.] Thus there's far more to the Internet's transport service (TCP) than JANET's (which is practically non-existant). TCP makes no assumptions about the underlying network whereas JANET leaves nearly everything for X.25 to sort out. Now consider the network interface to the operating system. In most cases, the network protocol processing is not the major part of the system's overheads. Other factors like getting data to/from the network interface, context switching and data copying are more important. I would suggest that TCP/IP offers much less overhead than X.25 here mainly because vendors will have invested much more effort in improving TCP/IP since its usage is likely to be far more widespread than X.25. [You cannot imagine Sun (say) expending the same sort of development effort on CB as they have on TCP/IP.] In some cases, this will happen automatically: almost all UNIX TCP/IP implementations derive from the 4.3 BSD code which is highly tuned. In terms of CPU cycles burned for a connection, X.25 probably needs less than TCP, but X.25 will have a less well tuned interface to the rest of the OS. The end result is that both require pretty much the same amount of system resources. Then, there's the relative merits of the higher-level protocols. In terms of overheads, there's not much difference between NIFTP and FTP. Both have strengths and weaknesses which tend to equalise things. FTP loses out if network ports are a critical resource (they shouldn't be) because it needs two connections; one for data transfer and one for the interactive session. Likewise the number of CPU cycles needed for an X.29/TS29 session won't be significantly different from those needed by a telnet server. Don't forget that telnet is functionally superior to X.29: it offers more for roughly the same overheads. In short, determining the "more efficient" protocol depends on where you measure it. Of course, you then have to weigh up the benefits (or not) of efficiency with the functionality that the protocol offers. Deciding that question is a matter of religion. As to running tcp/ip over janet, any reason you couldn't run slip over a coloured book terminal session ? Yes. The JNT wouldn't like it. What you're proposing is not very efficient: a better (and easier) option is to put IP datagrams inside X.25 frames. That too would upset the JNT since you'd be running something other than CB on JANET. Jim