Roger Scantlebury

Transcription

Good evening Good evening ladies and gentlemen... (the microphone is on) Thanks very much for coming on a Monday night to celebrate the UK's computing heritage. It's fantastic to see so many of you here and so many faces that we recognize So as you know the UK has a fabulous and long computing heritage but it's not something that has necessarily reached a wider audience in recent years, it's kind of a hidden secret very often and this is something that we at Google have spent some time quietly trying to encourage more knowledge about the UK's computing heritage over the last two or three years, so we're delighted to have you here tonight to meet some of the pioneers of the Internet... Thank you for not saying dinosaurs! To look at some some films that we've been producing over the last couple of years and to hear some some talk so the evening is in a couple of parts, we're going to talk here to the four musketeers of the Internet in a moment Then we'll have a little bit of break, an opportunity to mingle some more eat some sweets and popcorn and then we're going to come back with a couple films a couple of talks and a Q&A with Vint at the end of the evening right so I'm going to sit down and we'll kick off So let me introduce you to four of the great pioneers of the internet, starting on the extreme left there, Peter Wilkinson and Roger Scantlebury who were at NPL back in the 1960's and 70's and really helped develop I suppose what you could call the first Internet like thing in the UK and indeed in the world and certainly the term packet switching developed, or coined for the first time at NPL in the 60's Vint Cerf now obviously at Google, Chief Internet Evangelist at Google and Google glass wearer That would be me! who was at Arpanet at Stanford again back in the 60's and 70's actually at UCLA it was Arpanet Internet is Stanford, you know the difference between Stanford and UCLA we'll get to that.... is there a difference? we'll come to that and DARPA as well and Peter Kirstein University of London and the first person to bring the Arpanet node to University of London in the early 1970's I hope I made that introduction reasonably accurately So really, my role in this is not to intervene too much but really to wind them up and and let them go and having had a little bit of a conversation with them earlier, I don't think that's going to be too much of a problem So Roger, I was going to kick off with you... If you don't mind I'd like to grab a hold of this... he always does! ... it's habit, sheer habit! so um... Peter and Roger were both at the National Physical Laboratory and the then superintendent of the computer science division was Donald Davies who was instrumental in inventing the term packet, he worked with a linguist to do that but what was more important is that he initiated a project there to build what was intended to be a national scale packet switched network so the documentation, the design work and everything else all of which took place during the mid 1960's and to the late 60's and early 70s It was intended to be a project not at all unlike the Arpanet project which began in 1968 so I mean, they were essentially contemporary projects each of them intended to be a small-scale network to test the idea of doing store and forward packet switching so Roger and Peter were very very much a part of that early design work and in fact I'll let you guys tell part of that story ... Roger brought... was the first person to bring some of these ideas from the UK to the US in 1967 at a conference in Gatlinburg Tennessee where he met with Larry Roberts who was the primary overall manager of the Arpanet project for the defense department, so can you give us a sense for how this thing actually got started... in fact what the hell was NPL doing in communications when that wasn't your charter well that is true, and of course we got rapped over the knuckles for doing it but hey, what the hell you know Perhaps it would be a good idea to put it in context Donald Davies as you said was the superintendent of the computer science division and our computer science division had had a long history in computing going back in fact to Alan Turing who came to NPL in 1946 I think and in fact I've got a picture of the Colossus machine... there's Alan Turing he was responsible, well not responsible solely but was one of the prime movers and shakers in the Bletchley Park machine, the Colossus... Shall we pause for a moment, how many people have had an opportunity to go to the computer history museum in Bletchley Park? a few... quite a few... Good OK... So you guys will have seen this machine actually still operating which is a remarkable feat As I say, Turing came to NPL in '46 to start work on a computing project we were going to build a thing called the ACE computer after about a year he got fed up with the prevarications of the civil service who operated the national physical laboratory and decided to go and work with Max Newman at Manchester on the Manchester Mark 1 fortunately for us he left behind a dedicated team one of whom was Donald Davies who was his sort of immediate aide and some others like Ted Newman and the mathematician Jim Wilkinson not me, no, no relation! and several other noteworthy people and they built a pilot model of the ACE, they decided to do a cut down version first and this is a picture of the pilot model ACE which I think is actually in the Science Museum now... ...but not functional is that correct? I don't think it's functional ... [no it's in the Alan Turing exhibition] okay that machine first ran serious programs in 1950 and there's a bit more to it than actually shown in that photograph there was some.. the short term memory, the RAM was mercury delay lines, sonic mercury delay lines if somebody had walked into my office and told me they were going to build a memory system based on sonic transmission in a mercury delay line I would have thrown them out of the office this is nuts, how could that possibly ... strangely enough it worked they were kept in a special heat constrained room because the mercury would change length with temperature.. and then we had some drums to the machine for my sins I joined this team in 1955 would you believe and worked a little bit on this machine but on its commercial successor which was the the Deuce computer, I think we have a picture of the Deuce [there we are, that was it] and here we see hardware and software ... oh this is Deuce because the first one was ACE and the second... oh boy oh yes, minus two! yes so the software is in the punch card box here in the ... just in case there are any misunderstanding here that was a commercial machine, I think there were about thirty-odd of those machines actually sold in the UK so it was quite a successful machine... thermionic tubes of course ... So ACE was tubes also? ACE was tubes and the follow-on to that, the thing that was called the full-scale ACE was also tubes there were tubes for another ten years the 1794 of that era was the first one which was semiconducting but that's the background of computing in NPL and then in the early 1960's Donald Davies had become aware that the next big thing was going to be the communications between computers and had fairly quickly come to the conclusion that the telephone network was not the right vehicle for connecting computers together we had a meeting with the Post Office research people in Dollis Hill I seem to remember in the early 1950's ah, early 1960's I'm sorry and they the chairman was an erstwhile analog switched telephone network man, he had been chairman of the CCITT and he just gently pointed out to us that digital telephony was really nowhere there were millions of telephone handsets and about two thousand modems in existence, so clearly it wasn't going to go anywhere and we should stop messing about... and stop wasting time go away and do some thermometers or something or whatever NPL was supposed to do So Donald, of course once he became superintendent had a budget and was able to ... I think it's called Black Ops well, some of us in the US we call these skunkworks because no-one wants to go there because it smells bad and they stay away and you get some work done and that's where we started the work on the NPL proposal for a national network quite rightly so ...and that work started in 65 and we published a report in 1967 which I had the privilege of taking to Gatlinburg where I met Larry Roberts for the first time and the rest as they say is history ... History or disagreed history as one who tried to write a paper on that, and then had to try to make peace I'm not sure it's all history well one question i have for Mr Wilkinson is what was your role.. in fact, for the two of you what was your role in the project itself, the network project I was a sort of software guy you were the software guy, OK I'd been a part of the team working on the feasibility study and it came to the point where we decided we were going to build our own local area network within the NPL campus so the finger pointed at me Wilkinson you'll write the network software won't you so I said fine, ok, yes I'd expected to be asked but I really didn't have a clue where I was going to start so what computer did you use to build the packet switch at NPL? We chose the same computer as it happened that the Arpa people had chosen DDP-516 doesn't tell you a lot, does it really this is a complete coincidence right, because the Arpanet used the DDP-516 machines from Honeywell also instead of the Plessey machine We'd originally chosen the machine the Plessey XL 12 machine to do this role, and the reason we chose it was because it had a digital direct memory access system and you could attach 500 devices to the DMA channel which meant you could input and output to devices without interrupting sorry, without running through the processer, so the processer merely had to do buffer allocation and that sort of stuff... And it also had a good interrupt system which was pretty crucial I think It would turn out that the .. DDP-516 was similar almost as good architecturally and the Honeywell team were extremely collaborative and cooperative and would do what we wanted to bring it up to our requirements... If they actually built a ... we slipped them the idea of this DMA channel and they built it for us so within about, I think it was less than a year, about 9 months from the time we learned that we weren't allowed to buy the XL 12 because somebody in the Ministry had decided that Plessey didn't make computers they were just told, sorry you don't make computers and so we got the DDP-516 team to build the direct memory input output and if you go back to that picture on this side you'll see all the digital equipment, the input-output mechanism and the multiplexes and the line drivers that fanned out all over the the lab The National Physical Laboratory covered an area of 78 acres and there were.. ultimately we had several hundred terminals and computers attached to the network and they would all run through this digital network and in fact it was like a modern local area network... it ran at 1 megabit per second it was a 1 node network although the design was intended for multiple nodes but the idea was that this would be a packet switch and it indeed was a packet switch and we weren't allowed to build a wide area network because we were forbidden, so we just built the local area network because they couldn't stop us... so what's interesting is they used a T1 speed system a megabits based system in order to connect the devices to this switch and it would have connected the switches to each other In fact I'd like to comment on that particularly there were some other very significant differences in the thinking behind the NPL network and the thinking behind the Arpanet one of the problems which Arpa had at that time was that they were not in the. they have to they had to connect computers together and Steve Lukasik as director of DARPA spent many many anxious months trying to prepare papers to prove he wasn't in communications... because if it was the defence communications agency would have taken over the project all he was doing was connecting computers by contrast NPL started ... at least Donald started from the other side he said, look whatever the british might be saying about analog switching and one must remember that in around about '61 '62 one had the first digital telephone switches they didn't work very well was one problem, but much more important because of something which may sound familiar to you, because of problems with unemployment one decided to stay in analog switches for another ten years because the analog switches were built in Liverpool and Nottingham and shelve digital switches Donald however said look the T1 well, depending on where you are whether it was 2.048... it was 2.048 as far as the British were concerned, a T1 that is the future of communications whatever the British post office and DTI might say, therefore we should build this lowest level of switching into the same communication infrastructure that was a thinking which actually didn't occur in the US until quite a few decades later... that's true, that's true even though I'm of course a great believer in what you were doing of course... so this actually is a very interesting observation because the T1 notion arose in the telecommunications world more or less in the same period of time in the early 1960s and yet by the time DARPA was doing the Arpanet project it took convincing by Roger at this meeting in Gatlinburg to recommend that the Arpanet project which is intended to be a wide area four node network initially and then it would grow if it worked they were looking at 2.4 kilobit circuits using relatively low speed modems, at the time 2.4 kilobits was thought to be a moderate speed and Roger quite correctly pointed out and partly on behalf of Donald Davies that a much higher speed would reduce the delay for forwarding the packets through the store and forward network that convinced Larry to go to 50 kilobits a second but what puzzles me and i hadn't thought about it till now is that the T1 was actually functionally inside the network but it would not occur to the telephone network people to externalise access to the T1, that was their trunk speed what the hell do you think you're doing connecting your peripheral device at our trunk speed, you'd overwhelm the trunks in no time at all so we ended up with bonding 12 analog voice channels together to make 50 kilobits per second with a Bell 303 A modem that was half the size of a refrigerator but we chose the same DDP-516 computers completely independent of each other and built the Arpanet in 1968... you are aware that the guys from Bolt Beranek and Newman actually came over to NPL in I think, '68 for a kind of a... we had exactly the same problems you know, we weren't allowed by the record carriers to do wide area networks we solved it one way the Arpa people called it something else and solved it another way but a lot of that thinking was discussed ... that's interesting, I didn't know ... now the T1 carrier of course was invented as a way of multiplexing lots of speech channels over a pair of wires because they were running out of duct space in the ground and so to take a whole T1 carrier when you can actually multiplex 24 or 32 speech channels as you say, what are you doing, you know you can't do that, well actually you can so you were running 25 times faster than we were in the backbone but our problem was to build a wide area network you had to get channels that would go over a wide area and they refused to offer us anything more than 50 kilobits a second so now Peter Kirstein in 1973 receives the first Arpanet node in the UK connected at the time to the Norwegian defence research establishment by way of a 9.6 kilobit landline... I think I'd go back a little bit further first of all the one which was supposed to go onto Arpanet was the NPL what actually happened was that DARPA which i think that at stage it was ARPA, the advanced research projects agency was very advanced technically but completely and absolutely naive at what communications... international communications was about unsophisticated would be the best way of saying it they're a bunch of geeks right, they didn't know anything about telecom policy and any of you who have any idea of geography would understand why one might have three large seismic arrays in Norway Montana and Alaska it's purely arbitary of course where they were politically placed pointing somewhere... but they did happen to have three seismic arrays remember this is in the days of the limited test ban treaty and the question was had the Russians exceeded the limits of the test ban treaty, how could we tell... they wouldn't let us colocate with their test so we had to detect what they had done using the seismic array system and Norway was a convenient location... and that there was a treaty at very high-level organised between parliament and congress on collaboration on this seismic array business well, although on the whole we regard things the packet switching, as far as Arpa was concerned one of the very very important uses of this peculiar thing they were building was actually to connect things like these seismic arrays back... and that was computer connection after all back into some big data stores which they were building in Washington so somewhere around 1969 1970 when Arpanet was just starting Larry Robers had the mission of putting this seismic array onto the Arpanet and put it in with a 2.4 kilobit line well, by that time modems they were very expensive but you could just about go up to 9.6 kilobits shocking... this was possible and so obviously in those days there were almost no earth stations one of the very few places which had an earth station was Goonhilly and so the only way you could connect a norwegian earth station was to take a satellite connection to Goonhilly go by landline to London go by undersea cable to Norway and that way you put the seismic array on well I did start by saying that Larry was a little simple-minded in some things he thought well you just cut the line, it's already going to London just have it go to the NPL connect them onto the Arpanet too and then go on but there were two of three problems there were some minor problems about communications but I'll forget those because I have to solve those two what was rather more serious was that five years earlier Mon Général General De Gaulle had said no non to the british joining the common market this may sound familiar too but that's beside the point ... he had said no and so at this stage when it was suggested that the NPL go on to the Arpanet at the highest and I mean the highest level of government the british government said non orno because america was bad USA was bad, europe was good therefore... we've come full circle haven't we!... well I didn't like to say that! our papers say the same sort of thing!... but therefore Donald was not allowed to connect it on well I was definitely second best but I had had a certain awareness of the US and various things at the time so he gave me the largest amount that he was able to do from his own budget because as soon as he went outside his own budget there'd be minor political problems which would become bigger political problems which have become another no and so he offered me five thousand pounds to connect... that was when five thousand pounds was actually worth something you could buy a whole computer for that well, you couldn't, no, computers were expensive too so he offered me five thousand pounds and I would put in a proposal to the national research... to the science research council that with this offer from Larry Roberts so that went in but nobody in the government believed this kind offer from Arpa and the next that happened was a telegram went this was the days when telegrams were existing and the telegram went from the director of the research council to the director of DARPA requesting confirmation of this offer this was the first the director of DARPA had heard of a ... an international connection to a rather sensitive... not classified but sensitive installation inside funded by ARPA.... To make it even worse, you want to connect a college academic thing to our seismic array in Norway, are you crazy? it was worse than that I also wanted to find information about how a defense line LO 51 was connected to the US from Norway through Britain and those days academics were not allowed to know anything about defense lines didn't we run into a problem with the fact that you were a civilian university and Arpa was a US defense activity and the seismic array was a defense activity, and how can you connect a military thing to a civilian thing... oh it got a lot worse than that if you look at this picture behind you eventually when we were I can't remember how we actually worked the Oracle, I think you were a great deal to do with this Peter, but we actually managed to connect the NPL network through UCL to the Arpa network, and at the same time of course there was a thing called the European informatics network and NPL was the UK node on the european informatics network which was also connected to the french research network the Cyclades network and all of a sudden the post office and Cable and Wireless and AT&T became suddenly extremely scared that there was a big scale international network and all hell was let loose we had to sign disclaimers this long about what we were and were not allowed to do through this connection... so I want to take advantage of this part of the discussion to point out to you that if you think that computer networking is mostly about technology you're wrong. It's mostly about trying to solve bureaucratic problems and getting people to cooperate with each other. The policy problems are harder than the technical ones... well one of the fellows who was in charge at that stage of post office policy was somebody who you may remember, George Orchard and I was having lunch with him one time and of course in those days the post office had a complete monopoly on message switching and if you think message switching is something electronic, you're wrong if you do a semaphore across a public right of way that was probably message switching ...optical!... and things like emails had sort of been heard of and were clearly and strictly completely illegal so i put a problem it was illegal if you were connected to the telephone so I put a little problem to George I said George I have a computer and my computer has files and for 59 minutes well for 50 minutes I said first in each hour it's connected to telephone and during that time people put their data into my files then I disconnect the telephone network and do a little bit of sorting of the files and then I connect onto the telephone network again where have I done something wrong and when he had little difficulty with that, well what happens if I change it to 59 minutes and one minute or 59 minutes and 59 seconds and one second where was the illegality and I'm afraid his answer was oh Peter I wish you hadn't asked me that question so what what's your... one thing i would like to suggest now and I don't know how your timing is... I think we've got about five minutes, oh 10 minutes ok so another thing, if we could have that little diagram back again for a moment this doesn't show the NDRE component which would have gone up from UCL... It wasn't significant sorry?... It was not significant it was also part of a military system... what do you mean by it wasn't significant? what I mean by that, there was a very very important difference between what the norwegians did and what the british were doing the norwegian ... no part of NDRE the norwegian defence research establishment was attached only the seismic array ... there was no onward links from the large computer in Kjeller it was attached by itself and there were no links beyond it by contrast from the beginning the british we had several networks and if I may just say one more thing about that one of the biggest networks on it at the time was the Hindley Physics network and that included of course a link onto CERN and so of course I got a question from the beginning from CERN I had to ask for permission from anybody who was on it could they link in to the Arpanet so of course I had to ask my committee which included the British post office, could they and the answer was of course no but then they went on to say but how are you preventing it so I sent back a long answer there are the following six methods in which they could get on the following five I can prevent in the following way but the sixth, I could put up a notice if you do A B C D and E you would get onto the Arpanet but that's illegal shall I put up that notice? the answer was, no I shouldn't! so you can see, we weren't always popular the attitude actually at the Post Office office was we were completely harmless so let us do what we like and watch us and unlike these dangerous people like the National Physical Laboratory who would of course move the country we would do absolutely nothing which was of any risk to anybody so now I think I'd like to pick up the trace here in a couple of ways First of all, when the first node of the Arpanet was installed at your facility at university college london the first european not just the first ... the first european node the first node of the Arpanet was installed at UCLA in fact in 1969 but it's 1973 and this node is connected I think the way it worked was there was a satellite link that went from the east coast of the US internal to the Arpanet dropping down to an imp at Norsar in norway kjeller and then a landline from there went to UCL but we were interested in experimenting with packet satellite in addition to dedicated landline packet switching so not long thereafter a packet satellite effort was put into operation using Eutelsat 4A over the Atlantic then it linked the eastern coast of the US with the west coast of europe and the idea was to have multiple nodes in the packet satellite system sharing a common satellite channel, it was like an ethernet in the sky it had a bandwidth of 64 kilobits a second.. so these guys are still blasting along at a megabit and a half or 2 megabits and we're still farting along, you know at our little tens of kilobits a second eventually I think we doubled that to 128 kilobits but the concept was actually pretty interesting because these multiple nodes would share a common satellite channel it was a multi destination half duplex system and I remember when we ordered we put out a bid for a multi destination half duplex satellite system and we got back a number of bids and one from western union international read you know, dear DARPA we are delighted to provide you with a multi destination half duplex system, by the way what is it so we designed and built under DARPA for sponsorship this dynamic packet satellite system and eventually Peter's system was cut off from the internal Arpanet and forced to operate solely over the packet satellite network in the meantime in the united states we were experimenting with mobile packet radio which is what you carry around in your pockets these days at the time it required a cubic foot packet radio system that cost fifty thousand dollars per radio they were operating at a hundred to four hundred kilobits a second in the 1710 to 1850 radio band and they were using in that system spread spectrum CDMA now this is in the early 1970's so it was a very very advanced system for its time but the part I wanted to get to is that we've got these three networks now plus the ones that you see here all interconnected over satellite link and the packet radio system and the Arpanet Peter's job at UCL was to implement the internet protocols so his team was part of the first three groups that implemented TCP/IP starting in 1975 the other one was at Bolt Beranek and Newman and the third one was in my lab at stanford university so we were doing tests at the blinding speed of three packets per second over this multi network system we were also experimenting with packet voice and with packet video, video a little bit later in the late 70's voice in the mid 1970s this is the part that I wanted to take advantage of because we were testing packet voice and with a fifty kilobit backbone in the Arpanet you couldn't put a 64 kilobit voice channel through there very well so we used linear predictive code with 10 parameters as a way of compressing the speech from sixty four kilobits down to 1800 bits per second what you do is you send the ten parameters that represent the diameter of a stacked array of cylinders that are changing as the voice is speaking and then you excite that with a Formant frequency and you just send those eleven things forwarded over to the other side and they undo the compression and produce sound well you lose a certain amount of quality when you go from 64 kilobits to 1800 bits per second and the way it sounded when it came out the other end was like a drunken norwegian so I'm at DARPA by this time and it's now time to demonstrate the packet voice system to the guys at DCA, the defence communications agency who are you know all about circuit switching and everything so I wonder how was I going to do this and I remember thinking well wait a minute the norwegian defence research establishment was part of the test so we got Ingvar Lund to be the speaker first we had him speak through the ordinary switch voice system called Audobon then we had him speak through our packet voice system and it sounded exactly the same so we didn't tell the generals that everybody would sound this way as they went through this system so these guys, NPL, UCL and these personally and their colleagues were very much involved in the earliest stages of both packet switching and international networking and interneting in the mid 1970's and that story is not very widely appreciated here in the UK. Do you remember there was a paper that you, Alex Mackenzie from Bolt Barenak and Newman Hubert Zimmerman from the Cyclades network and I was the fourth author and it was an inter-network protocol that was a proposal we were trying to show that something that was TCP/IP eventually was something that you would use between the systems at the end of this conglomeration of networks none of which was the same inside none of which had the same interface but which could be connected together... interoperability was everything and Cyclades, I'm glad that we have this up here as well because the French had their network which I think Segal was the packet switched network and Cyclades was the larger scale system with the host computers in it what's important i think is that the Internet project was informed very much by its interactions with people outside the united states so we when you keep hearing that the internet was an american invention and everything else that's not exactly correct because Bob Kahn and i spent a lot of time with our colleagues here and on the continent exploring this stuff, and I'm just getting the shut up now signal so I'll shut up now So we've got up to the mid 1970's but I'm afraid we're going to have to stop it there but thank you very much indeed gentleman we had the rolling stones at the weekend and that was great but I think actually this was the gig to be at tonight and I think our gentlemen are ... doing rather better than the the rolling stones in terms of remembering it all and giving it afresh so we've run a tiny little bit over, what i suggest is that we give our panellists a big round of applause...