Will SONET Changes Change The Future?

It used to be that “optical networking” and “SONET/SDH” were pretty much the same thing. During the 20 years since the first standards were written, nearly all carrier fiber has been based on SONET and its international version, Synchronous Digital Hierarchy (SDH). But now there are some indications that SONET is changing radically.

The debate over the future of SONET isn’t really about SONET itself, as much as it’s about the tension between optics and electronics in networking. If SONET’s not in our future, then it may be that our notion of how routers and switches build networks may be obsolete, too.

Matching SONET With Newer Data Protocols

Pressure to change SONET has arisen largely from the fact that users and carriers are fast transi tioning to packet traffic. Packet flows are bursty in nature, and this creates inefficiencies when they’re mapped to fixed-size SONET paths.

In addition, SONET’s granularity doesn’t match up well with some emerging data protocols. For example, Gigabit Ethernet falls in between levels of SONET capacity; an OC-12 is 622 Mbps, which isn’t enough to carry GigE, but an OC-48 is 2.4 Gbps, which is way too much. Furthermore, GigE’s bursty nature can yield only 20 percent link utilization, which wastes more bandwidth yet. If we assume an OC-48 is used to carry GigE at 20 percent utilization, we’ve dedicated 2.4 Gbps for an average 200 Mbps flow. That’s less than 10 percent efficiency.

There’s also pressure on SONET from the perspective of cost- effective access. Gigabit Ethernet access lines from a carrier are far cheaper than SONET access; a GigE connection probably costs about half as much as even an OC-12, and provides more bandwidth.

But SONET is adapting to the pressure. Three specific developments seem destined to remake SONET and move the optical layer into traditionally electrical-network territory.

* Virtual concatenation (VC) is a way around the rigid SONET channelization, where only fixed-sized pipes were available and the entire bandwidth of a pipe had to be contiguously allocated. With VC, a connection can be created by combining a bunch of channels (STS-1s, or about 51 Mbps each) that add up to something much closer to the “right” data rate. Not only that, these individual channels don’t have to be contiguous, or even follow the same path. Virtual concatenation is thus something like inverse multiplexing.

Virtual concatenation can reduce SONET bandwidth waste significantly; a GigE connection that needed an OC-48 before (wasting over half the bandwidth) could now use what is called a “STS-3c-7v” that adds up to 1,050 Mbps, a nearly perfect fit.

* The next member of our SONET enhancement trio is the Resilient Packet Ring (RPR). Just what RPR represents is at least a semantic issue; some describe it as an evolution to the SONET physical layer and others as a replacement for SONET. It’s sort of both, a new MAC- layer protocol built on top of the SONET physical layer, but it’s a major step forward from vanilla SONET for packet-dominated infrastructure.

RPR eliminates all the unexciting TDM pedestrianism of SONET, but retains its important features, including support for legacy TDM flows and high-availability configurations with fast switchover. One carrier told me that RPR was “what SONET would have been if the Internet had been popular in the mid-80s.”

* Finally, there’s dense wave division multiplexing (DWDM). Wavelength multiplexing has added enormously to the capacity of fiber without a corresponding cost. As a result, the cost per bit has fallen sharply and carriers now worry a bit less about bandwidth efficiency. In addition, many equipment vendors have introduced products designed for “wavelength switching” at the core. This creates the notion of a “lightpath” or dedicated series of wavelength hops from source to destination that in a way mimics the old SONET notion of the “path.”

Wagging The Dog

The interesting thing about this evolution of the optical layer is that while it’s supposed to be driven by migration toward packet infrastructure, many people think the optical tail will end up wagging the electrical dog. This doesn’t necessarily mean photonic switching replaces routing, but it may mean that switching takes place down in the optical layer.

Under the concept of “protocol handling,” you decode the protocol headers at the edge, determine where the destination is on the network, and drop the packet into the lightpath/datapath/ concatenated path that’s going to an appropriate “off-ramp.” Along the way it might be shuffled around between fibers and wavelengths, but that’s all handled at the optical level. “Protocol handling” is the technical basis for the federal “GIGBE” project, which will provide a wavelength to every military base and represents the largest use of optics outside a carrier deployment; many of the RBOC optical network RFPs are based on this idea as well.

For site-to-site enterprise data services, the idea is to link RPR datapaths and long-haul MPLS label switched paths (LSPs) or wavelengths with CPE. This would mean, for high-bit-rate services, an optical infrastructure. Carriers are also planning to deploy RPR- Ethernet edge devices to drop GigE directly from the rings; the whole end-to-end connection would never rise above the optical layer in this model.

For the more “community” services like the Internet, content, application services and extranet VPNs, you’d still need electrical handling at the edge, because “community” services are multi-point and don’t scale with optical point-to-point solutions like RPR- Ethernet. Even with these services, however, the core routers could be replaced by optical handling. The on- and off-ramp devices for this new optical network are likely to bulk up into the half- terabit capacity range under the pressure of IP traffic growth.

We’re also likely to see an increased focus on how these “ramp” devices manage the opto-electrical interface and request/manipulate the various paths, frames and concatenations that will go into creating interior routes over the optical network. These new edge devices are also likely to be put under considerably more reliability/availability pressure in the future.

What Will It all Mean?

The new optical age will affect everyone, from enterprise end users to equipment vendors. One general impact will be to enhance the value of “big-ness.” A DWDM core with an RPR metro network and an Ethernet edge isn’t the sort of thing a venture-funded startup carrier is likely to be able to deploy. This is a major carrier game. Same is true at the equipment level; you’ve got to be a player big enough and with a long enough track record to be trusted in order to sell into this new framework.

What this will mean for users is harder to say. Carriers aren’t going to undertake a massive infrastructure upgrade for the purposes of driving down the cost of legacy services. The whole optical upgrade is predicated on the assumption that there is a new mass- market opportunity for IP services and also a high-margin market for application and content services. It’s the deployment of these services, and their financial success, that will fund the new network. Since these new services will be targeted at SMB and the consumer, they’ll offer pretty good price points for the enterprise buyer as well, and they’ll likely displace some legacy services.

Over time, though, the new infrastructure will give enterprise buyers lower-cost bandwidth even in point-to-point intra-company applications. Since future networks will be designed for efficient packet transport, packet services will see the best pricing. Look for what are now called “transparent local services” (formerly known as transparent LAN services) based on 100 Mbps Ethernet, GigE and Fibre Channel to offer at least twice the bits per buck as traditional SONET TDM services. These services will also be extended nationally and internationally, and will become the leased lines of the future.

What about voice and leased-line services? There’s a pretty good chance that as users move to packet data services, they’ll move to packet forms of all their current traffic, and these legacy strategies for connection will fade away. When they do, SONET as we know it will fade away too.

Conclusion

The impending optical shift seems to point to a future network that’s not exactly the converged Internet. A real electrical convergence on IP will happen only if application and content services become the real “services” of the network, pulling the enterprise customers away from point-to-point site relationships into distributed application relationships.

Could we converge the other way, down the stack to an all- optical network? Not as long as consumer broadband is the driver. Features that facilitate finding partners, discovering resources, managing connection bandwidth, etc. are essential if the majority of revenue comes from smaller buyers. These consumer-oriented features require electrical interfaces.

The whole convergence issue is probably moot anyway. Falling bandwidth cost reduces the economic value of convergence. It will always be a darling with the media (and analysts) but lower-cost, flexible optics cuts the heart out of the justification for harmonizing on a single higher-layer protocol. It’s the service that demands conformance, and it’s too early to tell whether the “service” known as the Internet will be t\he baseline for the future, or whether multiple service frameworks will emerge on top of the new flexible optical world. Either way, technology in the network is changing from the bottom up

Optical networking refinements diminish the value of convergence on IP

The new technologies favor big carriers and vendors over startups

Tom Nolle is president of CIMI Corp., a consulting firm that specializes in advanced computer and communications networks. He can be reached at tnolle@cimicorp.com, and he is a member of the BCR Board of Contributors.

Copyright Business Communications Review Jun 2004