I handed over TW $1,000 (U.S. $40) to the cab driver and checked into my hotel, asking for a 4:45 A.M. wakeup call. It was 9 P.M.. While waiting for my baked dried squid with chili sauce to come up from room service, I read about SEEDNET, half of the Taiwanese Internet.

As with many countries, the Internet was split among multiple ministries. Education ran TANET for the universities and the commercial SEEDNET was run out of industry. The namespace was split between the two groups.

What was interesting about SEEDNET (but certainly not brought out in the paper I read) was that SEEDNET used a block of 249 Class C addresses. SEEDNET was connected to JVNCnet, which in turn was connected to the NSFNET backbone.

Having a country take 249 separate Class C addresses was an interesting illustration of the problem of the IP address space. Class B addresses were being rationed because of exhaustion of that portion of the address space, but giving out multiple Class C addresses could put an additional strain on the routers, already working hard to keep up with close to 5,000 known network addresses.

At the time of my visit to Taipei, SEEDNET was physically connected to JVNCnet, but the SEEDNET world was not announced over the NSFNET backbone to other regionals. If you attempted to connect to SEEDNET from JVNCnet, the routers would know how to route the packet. If you tried to do so from another regional, packets would disappear.

The SEEDNET problem was certainly just a temporary one, but it showed the strains that were beginning to appear on the routing infrastructure of the Internet. Older networks, such as the MILNET, had even more problems.

The MILNET routers could only handle routing information for 3,750 networks. Since the NSFNET had over 5,000 networks, the MILNET administrators had to decide which routes to accept and which to ignore. The decision had been that all international networks that were not associated with peer military organizations were cut off inside the MILNET. The Australian AARNet, for example, could no longer communicate directly with hosts inside the MILNET.

Cutting off people who probably wouldn’t talk to you is certainly a rational response to the problem of saturating the Internet. The problem, however, was that this didn’t solve the long-term problem of scaling the Internet. The Internet was doubling every 7 to 10 months and there were projections that the routing table size could easily grow by an order of magnitude if nothing was done.

Address space resolution was a big issue at all the IETF meetings. One solution which was proposed was to do away with the distinction of address classes and instead delegate arbitrary blocks of addresses to the regional networks, which would in turn delegate blocks to their clients.

Using blocks of addresses and delegation on the hierarchy meant that, with suitable changes in the routing protocols, a single network prefix and mask could be used to refer to what otherwise might have been several hundred (or even thousand) entries in a routing table.

I fell asleep trying to figure out why it had to be so hard to put together an integrated global Internet. There was obviously a need for many types of networks: the day of “the” network had long passed. Yet, this diversity meant that the network was starting to fragment and splinter into subsets of connectivity.

The next morning, dozing on the airplane to Hong Kong, I was awakened by the flight attendant who presented me with a large box with the imprint “Name Card Holder” on it. Looking around, I saw all the other business class passengers stuffing their gifts into carry-on bags.

I opened mine to find a handsome slab of wood with a large blue and green ceramic duck mounted on it. Presumably, the business cards would be inserted in the beak. While this functional yet attractive desk accessory would certainly have made an impressive addition to my office decor, I felt that my seatmate might make better use of it. I handed my gift to him and he whisked it into his bag before I could change my mind.