grahamt@syma.sussex.ac.uk (Graham Thomas) (06/15/91)
(I sent the following to Michael by mail. He thought that it, and his reply, raised points that were worth a wider airing. So here goes..) Michael Padlipsky wrote: > > Graham Thomas-- > > I'd be so fascinated to see an explanation of your depiction of TCP/IP > as "SEMI-open" [emphasis added] that I'd even forego the pleasure > of debating whether the originators of the "Coloured Books" SHOULD > have realized [or, more locally appropriate, realised] that OSI > would be a very long time in coming, in exchange. > > cheers, map > ------- Ah. First I'd have to know whether you consider TCP/IP to be completely open or completely closed! Maybe I shouldn't have said 'open' and should have said 'semi-standard' or something. What I meant was, TCP/IP is available for lots of different machines, and is common in certain economic sectors, especially in education, (and not in others - here at least, SNA rules the roost in banking), but it's not been ratified by international standards committees. But do tell me why the creators of Coloured Book software should have realised, in the late 1970's, that TCP/IP would be ahead of OSI in 1991. I should say that I don't have any stake in this - I'm a user rather than a network designer. I do have a social science interest in how networks and standards are developed. My own view is that, given what they started from, the creators of JANET did well to build a unified academic network in the UK by 1985. The world has now changed, and some of the choices that seemed good at the time have proved to be unviable in the longer term. I'd like to hear other people's views. Cheers, Graham
PADLIPSKY@A.ISI.EDU (Michael Padlipsky) (06/18/91)
[I, of course, replied:]
Hi--
The "Open" in OSI was originally touted as being the antithesis of
Proprietary. "Anybody can do it" sort of thing. Thus, depending on
your view of the comprehensibility of ISO specs, TCP/IP is arguably
MORE "open" than OSI, and in no meaningful sense less so. (The
lack of blessing by an international standards organization is both
circular reasoning and, to some, a further argument in TCP/IP's favor
[though rarely in its favour, apparently (just a little Panatlantic
orthographical humo{u}r, there)].)
Assuming you've figured out which side I'm on, I'll touch briefly on
why I'd claim the Colourisers SHOULD have known better: by the late
'70s TCP/IP implementations were running whilst the only thing the
ISORMites had running was their mouths.
(Plus, rudimentary awareness of the pace of the international standards
process should have indicated that it would take quite some time for
OSI to overcome TCP/IP's chronological lead. Say, at least half-a-dozen
years as of the late '70s. That it's a dozen and counting by now could not
necessarily have been anticipated, but was at least implicit in some
technoaesthetic critques of the early '80s.)
For more on "ISORMites", and what somebody once called the "electropolitics"
of it all in a review, see M. A. Padlipsky, _The Elements of Networking
Style_, Prentice-Hall, 1985, and--lest I be accused of advertizing--
Marshall Rose, _The Open Book_, Prentice-Hall, 1990. Although the
conclusions do differ, both books do, in my view, adduce rather
similar evidence about the technoholy wars between what I call the
ARM (ARPANET Reference Model) and the ISORM (ISO Reference Model)
--or, more accurately, between the adherents of the ARM and the ISORM.
cheers, map
-------goldstein@delni.enet.dec.com (Fred R. Goldstein) (06/20/91)
In article <12694360385.18.PADLIPSKY@A.ISI.EDU>, PADLIPSKY@A.ISI.EDU (Michael Padlipsky) writes... >Assuming you've figured out which side I'm on, I'll touch briefly on >why I'd claim the Colourisers SHOULD have known better: by the late >'70s TCP/IP implementations were running whilst the only thing the >ISORMites had running was their mouths. Now that you mention it, I've been looking for somebody who remembers the date. While TCP/IP was being developed in the '70s and had some lab implementations, the ARPAnet itself was running NCP for quite a while, until they were ready to switch everything over. I recall that one day (well, probably a long weekend) it happened, and NCP was No More. Anybody remember the date? I think that was the last time a protocol simply died, kaput, just like that, never to be heard from again. Well, ANF-10 may have outlasted it... --- Fred R. Goldstein Digital Equipment Corp., Littleton MA goldstein@delni.enet.dec.com voice: +1 508 952 3274 Do you think anyone else on the planet would share my opinions, let alone a multi-billion dollar corporation?
pcg@aber.ac.uk (Piercarlo Grandi) (06/20/91)
On 19 Jun 91 20:49:21 GMT, goldstein@delni.enet.dec.com (Fred R. Goldstein) said: goldstein> While TCP/IP was being developed in the '70s and had some lab goldstein> implementations, the ARPAnet itself was running NCP for quite goldstein> a while, until they were ready to switch everything over. I goldstein> recall that one day (well, probably a long weekend) it goldstein> happened, and NCP was No More. goldstein> Anybody remember the date? Not so suddendly. Comer says that TCP/IP introduced to ARPAnet around 1980, and that gradually sites started dropping NCP based services and using TCP/IP, until in 1983 nearlly all sites had converted. I guess that if you really wanted a cutoff date it would be hard to say; probably one could say it was when more than say 80% (but any percentage greater than 50% would probably do) of ARPAnet backbone traffic had become IP based. Somebody should have such statistics around... -- Piercarlo Grandi | ARPA: pcg%uk.ac.aber@nsfnet-relay.ac.uk Dept of CS, UCW Aberystwyth | UUCP: ...!mcsun!ukc!aber-cs!pcg Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg@aber.ac.uk
pcg@aber.ac.uk (Piercarlo Grandi) (06/20/91)
On 18 Jun 91 03:24:16 GMT, PADLIPSKY@A.ISI.EDU (Michael Padlipsky) said: PADLIPSKY> the Colourisers SHOULD have known better: by the late '70s PADLIPSKY> TCP/IP implementations were running whilst the only thing the PADLIPSKY> ISORMites had running was their mouths. Not necessarily so; they had viable if fiarly rudimentary X.25 technology to show. At the end of the seventies one could observe: 1) a large scale production WAN, the ARPAnet 2) experimental or otherwise small scale X.25 technology 3) experimental or otherwise small scale Internet technology 4) papers and intentions about ISO/OSI The ISO people *did* have something going for them; there were X.25 operational networks after all. Now comparisons 1 vs. 2 and 3 vs 4 point out that for *comparable* (ARPAnet vs. "X.25", Internet vs. ISO/OSI) levels of technology you are indeed right when you say that: PADLIPSKY> (Plus, rudimentary awareness of the pace of the international PADLIPSKY> standards process should have indicated that it would take PADLIPSKY> quite some time for OSI to overcome TCP/IP's chronological PADLIPSKY> lead. Say, at least half-a-dozen years as of the late '70s. -- Piercarlo Grandi | ARPA: pcg%uk.ac.aber@nsfnet-relay.ac.uk Dept of CS, UCW Aberystwyth | UUCP: ...!mcsun!ukc!aber-cs!pcg Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg@aber.ac.uk
mrc@milton.u.washington.edu (Mark Crispin) (06/21/91)
In article <PCG.91Jun20160421@aberdb.aber.ac.uk> pcg@aber.ac.uk (Piercarlo Grandi) writes: >Not so suddendly. Comer says that TCP/IP introduced to ARPAnet around >1980, and that gradually sites started dropping NCP based services and >using TCP/IP, until in 1983 nearlly all sites had converted. I guess >that if you really wanted a cutoff date it would be hard to say; >probably one could say it was when more than say 80% (but any percentage >greater than 50% would probably do) of ARPAnet backbone traffic had >become IP based. Somebody should have such statistics around... How soon these young whippersnappers forget. Perhaps sometime I should dig out my "I SURVIVED THE TCP TRANSITION - JANUARY 1, 1983" button. I'll have to check my copy of Comer's book to see what he actually said, but let me assure you that the transition from NCP to TCP was done in a great rush, occupying virtually everybody's time 100% in the year 1982. *Nobody* was ready. On January 1, DCA had the ARPANET IMP software modified so that they would not pass NCP traffic. There was a hue and cry, and thus for several months afterwards a number of sites (both NCP-only and NCP/TCP sites which needed to be able to talk to NCP-only sites) had "reclama" to use NCP. It was a major painful ordeal, and one that those who experienced it are not eager to repeat. You can tell who we are -- the old farts who smirk knowingly when "ISO transition" gets mentioned. IMHO, however, it is likely that if DCA didn't do this, that NCP would still be the standard protocol. Now that there is no single agency (particularly a military agency used to using its teeth) that runs the Internet, I doubt such a thing could be done again.
mcdonald@aries.scs.uiuc.edu (Doug McDonald) (06/22/91)
In article <PCG.91Jun20161256@aberdb.aber.ac.uk> pcg@aber.ac.uk (Piercarlo Grandi) writes: >On 18 Jun 91 03:24:16 GMT, PADLIPSKY@A.ISI.EDU (Michael Padlipsky) said: > >PADLIPSKY> the Colourisers SHOULD have known better: by the late '70s >PADLIPSKY> TCP/IP implementations were running whilst the only thing the >PADLIPSKY> ISORMites had running was their mouths. > >Not necessarily so; they had viable if fiarly rudimentary X.25 >technology to show. At the end of the seventies one could observe: > > 1) a large scale production WAN, the ARPAnet > 2) experimental or otherwise small scale X.25 technology > > 3) experimental or otherwise small scale Internet technology > 4) papers and intentions about ISO/OSI > A more interesting date, to me, was in late 1973 (or was it early 1974). I proved that the then-existing technology was not so robust by crashing the entire Arpanet and about 30% of the computers on it (all the IBM 360s). Three times in one day. This resulted in a big guru fest in an attempt to figure out how it happened. Doug McDonald
dlynch@ISI.EDU (Dan Lynch) (06/24/91)
The magic cutover date was 1/1/83. No more NCP. Just TCP/IP. Some bureaucrat chose the date. Dozens of us system managers found ourselves on a New Year's Eve trying to pull off this massive cutover. We had been working on it for over a year. There were hundreds of programs at hundreds of sites that had to be developed and debugged. Never again in history will there be such a clean cutover. The transition from TCP/IP to anything new will be done very slowly. I think that is a good idea. Dan PS. I'm the one who had the "I survived the TCP Transition" buttons made up. Made 500 of them and passed them out to the pioneers all over the globe. Save those buttons, gang!
mckenzie@bbn.com (Alex McKenzie) (06/24/91)
Mark Crispin and Dan Lynch have already stated the official NCP to TCP/IP transition date. They have correctly stated the mad rush at the end of 1982 to make the cutover. For the historically curious, here are a few more facts. The first TCP implementation I am aware of was for the PDP-10 running the TENEX operating system; it was reported to be running at BBN in February 1975. By November 1975 this implementation had been involved in experiments with a Stanford University implementation for the PDP-11. The experiments disclosed protocol design deficiencies which resulted in a revision of the TCP spec. The first TCP implementation for UNIX was written at BBN; it reached operational status by November 1977. A TCP/IP gateway written by BBN for the PDP-11/40 was in use interconnecting the ARPANET and the Packet Radio Network (a DARPA project) by May 1978. The RFC announcing that the official cutover from NCP to IP would be 1/1/83 was published in November 1981, allowing more than a year for host organizations do do planning and implementation. There were TCP/IP implementations for most ARPANET hosts already running (in at least prototype form) by November 1981. About 1/3 of the 285 general-purpose ARPANET hosts accepted TCP connections in early December 1982, about 1/2 in early January 1983, and about 2/3 by the end of February 1983. The IMP (packet switch) code which would disable transmission of NCP protocol traffic was actually never activated, although it had been written, tested, and installed. Cheers, Alex McKenzie BBN Laboratories
malis@BBN.COM (Andy Malis) (06/24/91)
> ... let me assure you that the transition from NCP to TCP was > done in a great rush, occupying virtually everybody's time 100% in the > year 1982. *Nobody* was ready. To the contrary - we (the enforcers) were! :-) > On January 1, DCA had the ARPANET IMP software modified so that they > would not pass NCP traffic. I wrote the IMP code to make this happen. We had a configuration bit for each host, to allow or disallow NCP on a host-by-host basis, plus a master switch for each IMP. To enforce the cutover, we preset all the configuration bits ahead of time, and then just flipped the master switches at cutover time. > There was a hue and cry, and thus for several months afterwards > a number of sites (both NCP-only and NCP/TCP sites which needed > to be able to talk to NCP-only sites) had "reclama" to use NCP. We had a long list of hosts that were granted permission, for one reason or another, to continue to use NCP. As I recall, we had a good number of hosts on the list from day 1, and it grew quite a bit before it started to shrink. Eventually, the list grew shorter and shorter, until (probably eight months or so after the cutover) the last hosts had their NCP permission turned off. That particular code is now long gone from the PSN. However, upon reflection, it was probably the first protocol-based filtering ever built into a packet switch. It used the 1822 "link number" protocol identifier to determine which messages were carrying NCP. > It was a major painful ordeal, and one that those who experienced it > are not eager to repeat. You can tell who we are -- the old farts who > smirk knowingly when "ISO transition" gets mentioned. I'm looking forward to it already. Andy
malis@BBN.COM (Andy Malis) (06/25/91)
Alex, I would like to disagree with you on one point: > The IMP (packet switch) code which would disable transmission > of NCP protocol traffic was actually never activated, although > it had been written, tested, and installed. We actually did pull the trigger, as it were. We came in on New Years Day and turned on the filtering in the IMPs. It wasn't until the 2nd that most of the complaints started coming in. Before the cutover, we had run statistics on which hosts still had NCP packets coming in and out, and we also knew which hosts were officially allowed to run NCP following the cutover, so we had a good idea where the complaints would be coming from. Andy