The multiplicity of efforts is necessary because the upgrade to the Internet is no longer just about preparing Web sites for the 21st century. Rather, it's about turning the Internet Protocol into the unifying language of all broadband communications. The circuit switches that carry nailed-up voice calls on a global basis, enabled by a switched infrastructure that took more than a century to build, are expected to be replaced within a decade by a single packet-switched network that carries all the world's data, voice and video. The belief that IP will constitute a telecommunications Esperanto is becoming nearly as accepted as the belief that global warming is real; the only remaining issue is what to do about it.

The task of making sure IP can take on this added burden is no longer the sole responsibility of the original 1960s developers in the Defense Advanced Research Projects Agency or of the recently announced Internet2. This time, thousands of smart people at hundreds of private companies are taking on small chunks of the problem, working at all seven layers of the Open Systems Interconnect protocol stack. Their goal: Allow a single packet-switched network and a single IP protocol to provide all the world's citizens with instantaneous broadband communications.

At the largest scale, that of a global fiber backbone, it became obvious in the 1990s that a step forward would require a leap back into the past. Developers of the Synchronous Optical Network suite of standards had planned for fiber speeds that would reach 10 Gbits/second, 40 Gbits/s and beyond, but they had counted on the staged growth of the time-division-multiplexed transport of voice channels. When Sonet had to start carrying packet data traffic-and rather ineffectively at that-the math for Sonet broke down.

The answer of the early 1990s was wave-division multiplexing, in which one fiber would carry several voice circuits or IP flows, separated by color-of-light signal. Though WDM carried the panache of being new and modern because it was photonic, it represented a turn back to an analog world for a transport system that was supposed to go all-digital. The application of WDM to increase bandwidth was absolutely necessary for the spread of global IP transport.

What developers envision as a short-term transitional step, albeit with some potential for long-term usefulness in the backbone, is a shift in network topology from the metropolitan rings that grew out of Sonet research to a long-haul any-to-any mesh that is best for solving routing problems for IP traffic.

Sometimes the long-haul networks will be owned by one service provider, usually a specialized fiber-based interexchange carrier like Qwest Communications International Inc. or Level 3 Communications Inc. But even in the long-haul backbone, traffic could hop across several service-provider domains.

Although development activity can occasionally seem manic in the long-haul world, it pales in comparison with the action in metropolitan and regional networks.

Every local phone company has started its own Internet service provider (ISP) business. Don't look for an architectural winner in this space. Luminous Networks Inc.'s vice president of marketing Jay Shuler likes to use the old phrase, "Where you stand depends on where you sit."

When a physical network is a simple 10-Gbit Ethernet ring, or a packet-over-Sonet point-to-point link, the primary Layer 3 device can be the type of "core router" manufactured by Avici Systems Inc., Juniper Networks, Lucent's Nexabit group or Pluris Inc. Those routers can find a home in the peering centers located in both long-haul and metropolitan networks. But the closer a router gets to the traffic point of origin, the less its primary function is speed of transport and the more complex it must become in terms of aggregating traffic.

As part of an effort to create easily switched IP packet flows, Cisco Systems Inc. has proposed a model of appending tags to the IP packet, dubbed tag switching. This was standardized by the Internet Engineering Task Force as the Multi-Protocol Label Switching, or MPLS, model.

There are many new wrinkles left to explore at the network edge. In physical-layer transport, new optical companies are proposing passive optical network technologies to allow corporations to utilize dark fiber that extends directly to the corporate local-area network or the in-building service distribution point. There also are new modulation methods such as optical frequency-division multiplexing, subcarrier multiplexing and space-division multiplexing that may make it easier for a metropolitan aggregator box to offer several types of service at once. Ultimately, however, all metropolitan aggregation technologies are held hostage to the last-mile access problem, perhaps the thorniest of Internet evolution issues.

Despite significant progress in the digital subscriber line and cable-modem arenas-and several fixed-wireless access models look promising for mid-decade-universal broadband access is a concept that will take several years to implement, despite all the hype about the world turning to Internet time.

The emphasis on full IP interoperability has been important in the second half of 2000, as a backlash has begun against the lighter-weight Wireless Applications Protocol (WAP) promoted by Phone.com Inc.

While WAP had some early and vociferous adherents in Europe, the user-experience stories coming out of trial networks on the continent seem centered on the observation that "WAP is crap."

Given the number of investments made in wireless Internet services in the past few quarters and the subsequent anti-WAP backlash, it is safe to say that late 2000 and early 2001 may see a general anti-wireless Internet backlash that could further slow expansion of packet networks in this area.

Even with a working broadband client appliance, access to Web sites is only as dependable as the global caching and mirroring strategy employed by the ISPs between source and destination.

Strategies for handling caching and content replication have fallen into three rough categories. Inktomi Corp. came first with a centralized proxy-server concept, realized in network operation centers that served to mirror content as close as possible to the requester of the information. By working with service providers like Digital Island Inc., Inktomi has been able to assemble a reliable global caching service.

In early August, Inktomi joined with several vendors in promoting the Content-Bridge Alliance, in which content itself would be brokered in the same way that bandwidth was brokered in peering centers.

Others are following the route blazed by Akamai Technologies Inc., which actually creates a proprietary redirection vehicle for URLs, which are translated into Akamai-specific RLs for distribution through special accelerated subnetworks.

Finally, hardware-based vendors like CacheFlow Technologies Inc. are trying to place more caching closer to the end client, and make the Web servers do more in the process.

In the end, the lofty goal of a single packet-switched network and a single IP protocol for instantaneous broadband communications is still in the distance. But it should warm the Web user's heart to know that there are some smart Internet architects chipping away at the problems to reach that goal. The most notable are featured in this special issue.