Now, the advent of a series of Internet2 test beds informally referred to as the "bones"-Mbone, 6bone, Qbone-has given industry a peek into research the Internet2 community is performing in handling special-purpose traffic, and what the test results could mean to the riffraff managing the everyday commercial Internet.

Russ Hobby, a senior consulting engineer for Internet2 working out of the University of California at Davis, has been active in the Internet2 project since its inception in the mid-1990s. Hobby was one of the architects of the Bay Area Regional Research Network, or BarrNet, one of several academic regional networks funded by the National Science Foundation in the 1980s. This gave him experience working in the early Internet regional communities that formed the backbone of NSFnet, a precursor to the commercial Internet.

Hobby has been involved in the Internet Engineering Task Force for more than a decade, and chaired the working group that developed the Point-to-Point Protocol. He was the applications area director for the first Internet Engineering Steering Group, which helped drive the standard-setting process for IETF that was later used to develop a wealth of standards.

Internet2, formed by the University Corporation for Advanced Internet Development, or Ucaid, is quite different from the Next Generation Internet project, which has official federal funding through the Defense Advanced Research Projects Agency. Darpa's broadband information technology program provides official federal funding for academic and federal-lab test beds in three areas: ultrahigh bandwidth, including bandwidth on demand; network management and monitoring; and application performance.

By contrast, Ucaid recruited the assistance of corporations like Qwest Communications International Inc., Cisco Systems Inc. and Nortel Networks Corp., but never served as a direct conduit for federal funds.

Hobby helped form many of the working groups at the 1995 Monterey (Calif.) National Information Infrastructure Conference that defined the Internet2 Project. Though the Abilene backbone network was announced at the same time that many Internet2 goals were presented to the public, it was not a foregone conclusion that Internet2 would have its own dedicated network and specialized network nodes at various universities. Abilene formed the scaffolding upon which special test beds like Mbone (multicast backbone) and Qbone (Quality of Service, or QoS, backbone) were built.

The test bones were created out of working groups established within Internet2, though any members of the consortium at large could participate in tests.

Multicasting, in which traffic is sent out to defined groups of destination addresses, is an area where consortium members have the most experience, since Mbone has existed since 1996. The team has concluded, through some tough trial-and-error experiments, that multicasting protocols must be as simple as possible. Mbone participants have found scaling problems with some of the original multicast protocols, such as the Distance Vector Multicast Routing Protocol and Protocol-Independent Multicast (PIM), and now are centering on PIM-Sparse as a preferred protocol.

"This doesn't mean the industry has to follow our lead," Hobby said. "But the experience with Mbone suggests that it's advantageous to look at simple, lightweight multicast support."

Qbone is a newer entrant, meant to test both reservation-based "nail-up" services discussed in the Internet Engineering Task Force's Integrated Services working group, and the per-hop-behavior flow-aggregation methods handled in the Differentiated Services group. Reflecting the industry's own view on scaling, Qbone participants have spent more time looking at DiffServ, where aggregate flows of IP traffic can be sculpted, than the IntServ process, which requires independent reservations for each IP flow. At a presentation at the IEEE's International Conference on Communications last June, Hobby stressed that IntServ still may prove useful for creating circuitlike links of IP flows for certain applications, but that Qbone work suggests that DiffServ has wider applications, especially in large-scale commercial networks.

"We don't have religious views within Internet2 on IntServ vs. DiffServ," Hobby said. "We're looking at where different QoS methods make sense in different parts of the network. Even bandwidth overkill has its role in solving bottlenecks, though opting for an overkill cannot last you forever."

Hobby has been less involved in 6bone, a network established by the IP Next Generation, or IPv6, working group. Abilene as a whole began some limited IPv6 use in May of this year, providing the first opportunity to turn the early 6bone experiments into a national IPv6 test bed.