There are several questions crying out aloud for answers. For instance, high-speed networks will carry a combination of vast amounts of voice, data and video. This information must be compressed either losslessly or lossily in an effi- cient manner which is in conformity with network protocols. What are good mathematical models for such sources? Are statistics-based compression schemes the best choice? How do we characterize the fundamental limits of compression for such sources?

Another class of problems arises in wireless. Given that we're all expected to ultimately have cellular units glued to our persons, we surely have a personal stake in this area! I am told that the compound channel is too optimistic a model for the wireless link whereas the arbitrarily varying channel is too pessimistic. Is there then a better model with a capacity which can be evaluated? What happens to this channel in a network context? What are good coding and modulation formats for this application? The CDMA/TDMA debate has already been alluded to.

Next, there is a growing recognition of the fact that satellites will play a key role in networks which heretofore were considered to be largely terrestrial. The introduction of satellites raises several interesting issues, one of them being multicasting. Simply put, multicasting involves the transmission of the same information to many receivers with different reception capabilities. From an information theoretic standpoint, the problem is one of joint source- channel coding, with the source coding part being hierarchical in nature. There is also a (hierarchical) source coding problem with constraints being imposed by allowable network link rates. Yet another issue concerns the use of the satellite link as an alternative to terrestrial routes, e.g., when the latter are congested or fail. The inherent ``asymmetry" of the satellite link (owing to propagation delays and reduced speeds) in an otherwise terrestrial network raises a host of interesting problems.

Sergio has pointed out that multi-user information theory has not quite kept up with the development of networks. I do not know whether this is due to the fact that the problems, as formulated by us, are just too difficult, or whether new paradigms are needed. On a slightly different topic, Gallager -- in the late seventies -- published a paper on the information content of protocols. This paper had garnered much attention upon publi- cation, but somehow did not succeed in retaining it. As we are faced with the development of networks which will carry cells in an asynchronous manner, the information content of pro- tocols becomes increasingly relevant.

This broad area is clearly a rich source of hard problems for information theorists. Several studies are at present under way but the entire effort could do with some coherence.