Summary

  • MAE-East offered a shared Layer 2 path in the Washington area, but each network still chose which peers and routes to accept.
  • Its 1994 record shows the costs of concentrating traffic and route information in one growing exchange.

A network can share a room without sharing a rulebook. In the early 1990s, providers in the Washington region connected routers through MAE-East, a Metropolitan Fiber Systems exchange. That gave them a common physical place to meet. Whether two networks would pass traffic was still a separate decision.

The distinction is easy to lose in modern descriptions of an Internet exchange. A switch can put routers on the same local network; it cannot, by itself, make their operators announce routes to one another. The 1996 CAIDA history explains that exchange members could choose routes and filters bilaterally. Presence at an exchange did not guarantee reachability between every participant. CAIDA’s account places MAE-East in the Washington area about a year after the 1990 Commercial Internet Exchange, as an alternative interconnection point while the NSFNET acceptable-use rules still shaped commercial traffic.

A March 1994 Internet Engineering and Planning Group report captures the practical bargain. MAE-East’s extended Ethernet was carrying growing load. The report said packet loss peaked at 10–20 per cent, associated with collisions and long Ethernet segments in MFS’s configuration. Subscribers were working on performance while trying to preserve the Layer 2 exchange model. The same minutes describe the full routing table as too large for routers with 4 MB of memory; the MAE-East ICMNET routers had been configured with 64 MB. Route flaps were producing 30–50 updates per second in that setup. These were operators’ measurements and assessments recorded at the time, not an independent survey of every participant. The IEPG minutes also considered a geographically extended exchange toward Stockholm, but record it as a possible design, not a completed connection.

The later National Science Foundation NAP programme formalized a related transition. The 1993 solicitation separated the roles of backbone provider, exchange manager, routing arbiter and regional networks. In 1994, the IEPG described NAPs as Layer 2 facilities where policy would remain expressed through bilateral peering agreements; inter-NAP connectivity would depend on transit providers. The architecture kept the exchange from becoming a universal routing authority. The solicitation and IEPG record show that division as a design choice, not an automatic property of Ethernet.

By January 1997, one NANOG participant complained that the number of pairwise negotiations and the difficulty of finding the right contacts made peering slow. The post is one operator’s experience, but it reveals the other side of bilateral control: networks retained choice, while coordination carried a cost. The archived discussion did not argue that an exchange should force universal peering; it described a need for better information about who was willing to connect.

MAE-East’s historical importance is therefore not that it made every network one network. It placed independent routers within reach of one another at a shared interface. The exchange could reduce the physical work of interconnection while leaving route selection, commercial terms, and onward transit with the networks that joined. That boundary made local cooperation possible without handing the room authority over its participants.

Sources