LINX was having a real crisis of confidence in the year 2000. As the largest Internet exchange in Europe, it should have been confident about the future – but that was not the case. So, what was the problem?
The London Internet Exchange (LINX) had been established in 1994 as a place for ISPs to exchange their traffic. As one of the first such exchanges in Europe (not the first, that distinction probably goes to DGIX in Stockholm), it was established as a membership organisation, owned by the members that it served. It is a good model, probably the best for an Internet exchange, but we will return to that point later.
The establishment of LINX followed the precedent set in the USA, where the DARPANET had ‘leaked’ out of the military and academic sectors into commercial and domestic use. As the number of active ISPs multiplied, maintaining connections between them became an increasingly difficult problem to solve. A complete network of one to one connections simply does not scale, especially across such a vast country as the USA. Even though customer supplier relationships simplify this problem, what is needed is a mechanism for one to many or many to many connections. Out of this was born the concept of ‘network access points’ (NAPs), four of which were established in the USA by NSFnet – both to address the geographical problem of where to exchange traffic, and to enable the efficiencies of interconnect that were so vital to the growth of the early Internet. The interconnection process is called ‘peering’, where two networks agree to exchange traffic destined for each other. Peering (which is usually, but not always, settlement free) is the alternative to paying another network to ‘transit’ their traffic to and from the desired sources and destinations. In the early days of the Internet, the cost savings alone established a strong business case to seek to peer as much traffic as possible – to which you could add other significant motives in terms of control of routing and achieving low latency connections.
Unfortunately, these early interconnect arrangements did not end well, and the NAPs established in the USA were not successful in even the medium term, never mind the longer term (the original USA ones such as MAE East and MAE West do not exist anymore, in sharp contrast to those established around the rest of the world). This is quite a story in itself, and I will try to tell that in a future article if there is demand.
Returning to the LINX story, the crisis of confidence concerned the strong growth that was occurring in both the number of interconnected networks and the traffic flowing across the exchange. It is fair to say that the equipment that had been employed in the early Internet exchanges (I will use the shorthand descriptor IXP from now on, as the term NAP has largely dropped out of usage) were quite crude layer two devices. Much could and did go wrong, which is unsurprising considering that IXPs are classically heterogeneous environments – where ‘packets of death’ could be freely passed around to the potential downfall of all. Of course, I do not literally mean that, although the fear of it was very strong. There is actually quite a large range of things that can and did go wrong in the early IXPs, and many of the risks still exist today given the general fragility of Internet protocols.