Amsterdam was 0° C, certainly a change from Bangkok. I made a mad dash to the taxi stand and got to my hotel with plenty of time for a one-hour nap after my all-night flight.
At 1 P.M., I took a tram to Muiderpoort station, then got in a minibus that shot over a narrow little path running next to a canal and dropped me off at the complex of buildings holding Dutch research laboratories including the Centrum voor Wiskunde en Informatica (CWI), the computer science and mathematics institute, and the Nationaal Instituut voor Kernfysica en Hoge-Energiefysica (NIKHEF), the Dutch institute for atomic high energy physics.
NIKHEF and CWI have both been prominent centers of networking in Europe for many years. NIKHEF was the Dutch HEPnet site, CWI had given an early home to EUnet, and this complex of buildings on Kruislaan road kept popping up in one context or another. It housed the RARE secretariat, it hosted a gateway to IXI, and the president of RIPE was Rob Blokzijl, a NIKHEF employee.
It was to get to the bottom of all these projects and acronyms that I had come to Amsterdam. I had heard more and more about the maze of conflicting institutions, consortiums, organizations, and associations that made up the confused politics of European networking.
I was just in time for the eleventh meeting of Réseaux IP Européens (RIPE), an informal consortium of the TCP/IP research networks of Europe. I figured that RIPE must be something like the IETF, and thus a good place to gather information.
The meeting was about to start, so I gathered my badge and one each of the dozens of handouts and went into a university-style lecture hall. Down at the bottom of the hall sat three serious-looking RIPE officials facing a room of 50 or so equally serious-looking engineers.
Rob Blokzijl was going through a long agenda, reviewing action items from the previous meeting and making announcements. I propped my jet-lagged eyes open with toothpicks and started leafing through the materials I had picked up.
RIPE was formed as a sort of anti-organization, a reaction to the total ineffectiveness of other groups in setting up a pan-European Internet. At the time RIPE was formed, there had been several years of thrashing while people tried to figure out how to make OSI into something real.
Meanwhile, several organizations with work to accomplish had been putting together TCP/IP networks. An Internet was being formed, but on an ad hoc basis. Networks were touching and accidents were starting to happen.
“What kinds of accidents were happening?” I naively asked Rob the next day while we were hiding in his office during a slack time in the working group sessions.
“Routing loops, black holes, the usual,” Rob said. To try and solve the problems, Rob had convened a group of six people from the real networks in Europe. Nuclear physics was the lead player, so in addition to Rob there were representatives from CERN in Geneva and the Italian physics community. The Italians were operating the only 2 Mbps international link in Europe at the time. To supplement the physicists, representatives from NORDUnet and EUnet also came.
At first, the six tried to work through the Réseaux Associés pour la Recherche Européenne (RARE), the association of research networking groups in Europe. RARE wouldn’t touch this issue, though. The six researchers were operating TCP/IP networks and RARE was exclusively and officially an OSI group.
Borrowing a phrase from Daniel Karrenberg of CWI that the “time was ripe for networking in Europe,” an extremely lightweight organization was founded in September 1990 with the mission of helping to ensure cooperation among European IP networks. Rob Blokzijl formally announced the formation of the group in RFC 1181, a very carefully worded one-page document.
In Europe, careful wording is crucial and I’ve seen more than one group spend several hours wordsmithing seemingly innocuous phrases. Rob proudly pointed out the triumphs in the RFC, such as the sentence that “RIPE serves as a focal point for other common activities of the participants related to IP networking.”
Apparently meaningless to the untrained eye, this sentence is rife with subtleties meant to steer a course through the shoals of politics. RIPE is a focal point, not an operational unit. It worries about only its own voluntary participants and is not trying to impose solutions on others. It deals only with IP networking and is thus not advocating against OSI but merely serves as a forum for those who have already chosen.
By the time of the eleventh meeting, RIPE had grown to nearly 30 organizations and was meeting quarterly. NIKHEF hosted many of the meetings and acted as the organization’s secretariat. In terms of developing new standards, RIPE was a pretty sleepy forum. Those wanting a marathon of 50 working groups producing mountains of paper were directed instead to the IETF. RIPE was much more collegial and informal.
The week before the meeting, two days had been set aside for informal tutorials to bring new participants in TCP/IP up to speed. At the same time, a bake-off had been held featuring equipment from various vendors and even including an ISDN switch. People worked together to form different network topologies and see what types of problems arose.
The meeting itself was a one-day plenary, followed by a day of working groups, followed by another day of plenary meetings. There were only a few working groups, devoted to issues like routing and European connectivity.
In addition to representatives from most of the central European countries, there was a heavy contingent of East Europeans, including delegations from Czechoslovakia, Poland, Hungary, and Yugoslavia. Most of the representatives from Eastern Europe migrated to the working group session on connectivity. After a brief review of who was coming on line, the conversation turned to problems of mutual interest.
One of the biggest issues was simply getting equipment. All the countries were on the U.S. export restrictions list and the Poles told how it had taken them a year just to get a Cisco router. Various suggestions were bandied about on just exactly how the U.S. wanted the myriad forms filled out, but it was obvious that the rules the U.S. imposed remained mysterious and arbitrary.
Another working group was devoted to the issue of routing coordination. Routing was the single biggest technical issue at the meeting, and it is in this area that RIPE made its biggest contribution. In the U.S., there was a single backbone which used the EGP protocol to announce available routes to the second-tier regional networks. The backbone was centrally managed, so all the regionals got fairly consistent information. In Europe, there was no backbone, so it was pretty much guaranteed that conflicting routing information would be propagated, leading to routing loops and black holes. RIPE took a look at the map of Europe and it became clear that four sites acted as the main hubs.
These four sites, in Stockholm, Amsterdam, CERN, and Bologna, served as the centers for large numbers of secondary networks. Stockholm was the center of NORDUnet and Bologna was the center of the Italian network. Amsterdam and CERN had huge numbers of links going out to different countries.
The four sites agreed to take the lines running from Stockholm to Amsterdam to CERN to Bologna and turn them into a de facto European backbone. The routers at the four sites were combined into a single routing domain, thus presenting a consistent view of the world. To make the backbone a reality, all four sites agreed that transit traffic could traverse their links.
This was certainly not a centrally managed, fault resilient, highly designed backbone, but it did the job of connecting the European Internet. Since the four main sites also had links to the U.S., the European Internet was well connected to NSFNET and the rest of the world.
In a centrally managed backbone environment like NSFNET, low-level networks that wish to make their availability known to the world (and thus enable communications with said world) simply have to inform the NIC or NOC running the backbone. If the low-level network doesn’t fill out the appropriate forms or jump through the appropriate hoops, the network is simply not announced and thus is effectively isolated. It would be as if a road existed into a country but there was no indication of the road on any map.
In Europe, maintaining the name space was a trickier situation. RIPE came up with a novel plan to keep information up to date. A European WHOIS service was put into place containing network names, domain names, IP addresses, and the like.
The incentive to keep this information up to date was quite simple. The four backbone sites used the WHOIS database to load their routers. Sites that didn’t keep information up to date were simply not announced over the de facto backbone.
RIPE was a simple, effective tool for coordination and instruction, all done on a shoestring. The informal nature of the group and the advocacy of the heretical TCP/IP was too much for the networking establishment and RIPE came under heavy attack at first from officially sponsored groups like RARE.
To understand why RARE would object, one has only to remember that TCP/IP continues to be referred to by many as “DoD IP.” In contrast, OSI, developed with very heavy participation from European academics and PTTs, was the open way to do networking.
While RIPE was under attack, it was also clear that IP networking was gaining a strong hold in Europe. In one of those bizarre twists of European politics, RIPE became an official RARE “activity.” Not an official working group, mind you, with the connotations of European Commission financial support, official voting by one participant per country, and a chairman appointed by the RARE Council of Administration. Simply an activity.
Under the merger and acquisition agreement, RIPE would continue as an independent group and would elect its own chairman. RARE got to bring the renegades under its umbrella and unity once again reigned, at least on the surface.
To tell the truth, I wasn’t exactly sure why the merger took place, but I did get a glimpse when Rob told me about the new RIPE NCC being formed. The informal RIPE backbone had been working well, but members decided they needed a more professional Network Control Center. Not a mere NIC, several people cautioned me (but certainly less than a NOC), this NCC would serve a coordination and troubleshooting function.
The NCC would be funded by voluntary contributions. Since RIPE was not a legal entity, the money would be funneled through RARE. RIPE/RARE had already put out an RFP for ECU 250,000 (U.S. $307,500), enough they hoped for a manager, two staff, some equipment, a travel budget, and some office space. Obviously, it was hoped that a group like NIKHEF or NORDUnet would respond with an offer within the budget, and would donate resources on top of the ECU 250,000. A committee had been formed to evaluate the responses to the RFP, consisting of two RIPE officers and two RARE officers. This group, upon making its decision, would forward it to the RARE Executive, which in turn would make its recommendation and send it on up to the RARE Council of Administration. RARE would then officially hire (and fire) the manager.
Following this discussion, I must admit I was still a bit confused as to what was going on. While I wasn’t clear on the organizational ins and outs, it was certainly clear that RIPE was performing a valuable coordinating function, allowing the technical people to gather in one room and work out solutions to problems.
Sometimes the issues on the table were highly complex, such as how to do policy routing in an environment characterized by too many policies. Often, though, the issues were painfully simple.
Take the issue of country codes, the two letter abbreviations by which you send a letter or provide the root portion of a DNS name. The list of such codes used to be fairly static, but with the disintegration of the Soviet Union there were a bunch of new roots to the name tree.
Managing that root was a sign of power, and in more than one case, fights had broken out within a country between groups who felt they were the rightful namespace administrators. The right to control the top of the tree was a touchy issue in a place like Yugoslavia. Imagine a Serb administration of the Croatian subdomain, for example.
RIPE served as a place where these types of issues could be ironed out. Not everything can be solved on an informal technical level, of course, but it is amazing how many problems can be worked out this way.
One exchange in the RIPE plenary that I found interesting was the issue of where to find the current country codes. Officially, these codes are listed in an ISO standard and, in response to a query from the audience, a bureaucrat got up and explained that one obtained ISO standards from one’s own national standards body.
Not a very useful solution to a simple problem, so somebody else got up and suggested that the information be obtained from the back of a Sun manual. The representative from Israel suggested that Larry Landweber maintained this information as part of his list of connectivity.
Finally, somebody wanted to know if this information wasn’t online. He had heard about some standards server “someplace in the U.S.,” but when he tried to obtain standards there he had received an error message.
I had the dubious honor of trying to explain why the ITU had withdrawn their permission for posting standards since the server wasn’t helping “the right sort of people.” Many of the Eastern Europeans were furious about the decision since it had been their only access to the vital documents. Would ISO be posting their standards, somebody wanted to know?
I explained the concept of a standards haven in some friendly location like Budapest. Everybody laughed and the plenary broke up.
That afternoon, I managed to drag Glenn Kowack away from the center of the vortex he creates to find out about EUnet. Glenn had just finished his first year as the CEO of EUnet. EUnet was sort of the analog of the U.S. USENET, but had evolved into a much more structured network.
In the U.S., the UUCP protocols had long been used to allow people to form a very loose confederation of systems, a sort of network anarchy. A new system would join USENET by setting up an arrangement with some existing system for the transfer of mail and news, usually using dial-up links. The only real rule was the unwritten tradition that if somebody was nice enough to let you join, you ought to do the same for somebody else. In this way, a loose web formed across the U.S. and a de facto network came into being.
The informal nature of USENET meant no acceptable use policies or votes to allow a new member in. Lack of formal structure, together with the low cost, meant that USENET (and the underlying UUCP protocols) spread quickly among small startup companies.
Much of USENET ended up falling under the umbrella of USENIX, the U.S. UNIX users group. In Europe, simple networks were also forming in the context of UNIX user groups and the UUCP protocols. In Europe, though, there wasn’t quite the luxury for anarchy that the U.S. had.
For one thing, calls across national borders were expensive enough that it made sense to concentrate traffic through a star configuration of national nodes and a central switch. The central switch was the famous mcvax (now mcsun), housed at CWI in Amsterdam. Since an international call to one point in Europe costs pretty much the same as to any other point, there was no need for a fancy mesh topology, and the star stayed in place.
Because calls cost so much, users started paying their share of line charges from the very beginning. Administration and management was totally volunteer and fell under the umbrella of the European UNIX Users Group (EUUG), and the net came to be known as EUnet.
By the time Glenn joined EUnet in January 1990, it had grown to be a TCP/IP and WCP network with over 1,600 sites. Over the next year, it was to grow to well over 2,500 UUCP sites, 100 IP sites, and 750 news sites. All this traffic was going from local networks, run by a local UNIX user’s group, into a national node, and then over to Amsterdam.
EUnet, with all these users, began playing a major role in European networking, pooling its money with other groups like NORDUnet so that bigger pipes could be leased. Many of these shared lines formed the de facto European backbone of RIPE, and then later of EBONE.
One of the interesting aspects of EUnet is that it is often the very first network in a country. By January 1992, EUnet already had 22 national members, each representing a national UNIX user’s group, running a national node, and paying for their own line to mcsun. The cost of international links, the Amsterdam equipment, and miscellaneous overhead like Glenn were also apportioned among the members.
With the craze towards open systems at a peak, EEUG changed its name to EurOpen to signify openness, truth, and beauty. Thankfully, EUnet did not become OpenNet.
Both parts of the EurOpen name are applied in a fairly loose fashion. Open included things like UNIX and TCP/IP instead of being a code-word for only OSI. European is also interpreted kind of loosely. Iceland, Tunisia, and Moscow are members of EUnet, stretching the word far beyond the confines of the European Community, the European Free Trade Association, and even the European continental boundaries. Egypt and Algeria were coming on line soon and Glenn didn’t really seem to care who joined as long as he could provide service and spread networking into new countries.
In fact, it was the desire to spread networking to those who could use it that got Glenn his job with EUnet, in a very indirect fashion. In the 70s and 80s, Glenn led a reasonably normal life, taking jobs like that of an R&D manager for Gould in Champaign-Urbana, Illinois. Of course, he did take a few years out to help found and run a community radio station, a job that many engineers don’t sport on their resumes.
At the tail end of his stint with Gould, Glenn ended up as the first employee of the newly formed UNIX International. He enjoyed the job so much that, after the organization was off the ground, he quit and took three months off to go look at computers in Asia, visiting places like Hong Kong and Singapore.
He came back from his sabbatical and set up shop again in Shampoo-Banana, this time hanging out his shingle as a freelance consultant. He was doing various consulting jobs when Eastern Europe collapsed.
Glenn bought a Zenith laptop, packed his backpack, and took off. By the time he looked up, he found himself in Warsaw chairing the first meeting of the Polish UNIX Users Group, an event made all the more challenging by his lack of any knowledge of Polish.
In his travels, he made enough visits through Holland that he got to know the Amsterdam crowd well. EUnet decided to hire a real manager, and Glenn certainly seemed not only highly qualified but highly available. He took the job.
EUnet is proof that networks can be set up with minimal outlays of funds. With a headquarters budget of ECU 600,000 (U.S. $738,000) and a total pan-European outlay of well under ECU 10 million per year, EUnet has become one of the key European networks, acting as the lead player in bringing new countries into the world Internet.
EUnet has also played a key role in helping to form the EBONE consortium. EUnet’s historical role in working with other organizations on line sharing has meant that lines to the U.S. and across Europe have been possible even when groups like RARE were unable to act.
One would think that a low-budget, technically astute, extremely effective network would win plaudits from all over, but EUnet has its detractors. 05I-only groups view EUnet and its informality as a definite threat. In some cases, active efforts are underway to try and prevent EUnet from operating.
A recent example was brought to light at the RIPE meeting. SWITCH, the national research network in Switzerland, had barred EUnet traffic from its network. Not only had the traffic been barred, it was cut off without any advance notice to EUnet, causing several days of confusion as people all over Europe tried to figure out their sudden loss of connectivity.
At the RIPE plenary, Michael Nowlan, the Irish chairman of EurOpen, read a prepared statement protesting the lack of notice. What had really happened illustrated the type of politics that EUnet encountered.
In each country, it is up to the national UNIX users group to figure out how to move traffic from the national node out to the rest of the country. In some countries, a simple dial-up star configuration is sufficient. In others, there is enough traffic that a backbone linking regional hubs makes sense.
How to provide that backbone is up to each group. In some cases, the group will simply lease lines and set up a backbone. In other cases, line-sharing agreements with a national research network might be established. In other cases, a contract can be set up with some other entity to provide the backbone.
In Switzerland, the Swiss UNIX group (CHWG) had been using SWITCH, the national research network, as a backbone. CHUUG and SWITCH got into a squabble over various matters and CHUUG decided to set up its own backbone. SWITCH set up their routers to filter out IP traffic, thus cutting off connectivity from the 22 EUnet countries to anybody on SWITCH or with a tail link off of SWITCH.
Fortunately, these childish games are the exception rather than the rule. In most countries, EUnet happily coexists with the national research networks. The national nets handle the universities and the UNIX group handles low-volume UUCP clients, IP links to small companies, and anybody else who falls outside the official domain of “research” as defined in that country.
With my work for the day done, I went out for a couple of beers with Scott Williamson, the manager of the new Network Information Center (NIC) in the U.S. The NIC had been recently transferred from SRI to GSI, a subsidiary of Infonet. GSI and Scott had a rough few months as things got going and Scott was now making the tour of conferences like RIPE and the IETF to convince the community that things were back in control.
Running the NIC is no easy task, I learned. Just maintaining the WHOIS database requires an 800 Mbyte database running Ingres, a condition specified by the Defense Communications Agency, the official funding agency for the NIC. Even though a great part of the traffic into the NIC comes from the NSFNET, not the MILNET, the NSF had yet to set up its own similar facility, making Scott responsible for both networks.
One of the big problems was that the NIC had inherited its communications facilities as part of the DCA contract, which specified a 64 kbps link to the Internet. A pipe of that size is not enough for an organization that is at the center of the Internet and the link was quickly flooded. Only after the NASA Science Internet stepped in with a T1 line did things start to get better.
Over beers, I learned that Scott is ideally suited for his job. His background was managing a team of UNIX programmers for various Washington tasks, such as a job for the Federal Emergency Management Agency (FEMA) simulating telephone disasters. More to the point, though, he is a second-degree black belt in a Korean martial art and spent three years in the Australian outback. Both experiences breed qualities that are useful when dealing with Washington bureaucrats.
That night, RIPE had set up a group dinner at a local Indonesian restaurant. Indonesia was a former Dutch colony and an Indonesian dinner provided excellent food, at a price that was affordable on all manner of budgets, an important consideration given the wide geographical representation at the meeting.
Over salads in gloppy peanut sauce, preserved eggs in sweet chili and lemon grass, and fiery beef curries, the conversation drifted in and out of shop talk. I was too jet lagged and burnt out to think about networks, so I concentrated on the food.