Convergence is a dominant theme in modern telecommunications and networking, and it touches virtually all levels of the communications hierarchy. In the core, carriers desire a simple, unified network over which to offer largely packet-oriented services. For access, cable, DSL, and Passive Optical Network (PON) alternatives compete for the last mile, while the aggregation equipment transitions from ATM circuit uplinks to IP packets. In wireless, base stations are making a similar transition from ATM to IP. And in the enterprise, businesses wish to unify distinct voice and data systems in favor of a single, converged network.
This article will consider some of the history of the communications and data networking industries with an eye to how the disparate past is evolving into a unified future.
The past forty years of communications and networking industry development has left a vast footprint of technology standards and protocols. The earliest of these was voice, where the T-carrier system of the 1960's introduced the first digitization of speech. As these networks expanded and demand arose in the 1980's for data networking alongside voice, pressure developed for a standard interface between telecom operators as well as a roadmap to higher-bandwidth links; the result was the SONET protocol.
Simultaneously in the 1980's, Token Ring and Ethernet battled for LAN interconnection supremacy. As Ethernet achieved dominance by the mid-nineties, the vision of a unified network began to take hold among enterprises and service providers alike. For the enterprise, this was a means of dispensing with a separate phone network; for carriers, leased lines would become a thing of the past.
The telecom community produced the ATM standards in the 1990's as a solution to this unification problem, and this was an early step towards convergence. However, ATM was caught on the wrong side of a trend--the world was moving away from circuits and towards packets. Although ATM achieved wide deployment in many telecom verticals, particularly wireless and DSL aggregation, it never succeeded in the LAN; the cost advantages of Ethernet-based packet networking were too dramatic, and the applications at the time couldn't benefit from ATM service guarantees. By the late 1990's the prevailing view of convergence had shifted, and methods to scale Ethernet into traditional circuit realms gained momentum.
Today we are somewhere in the middle of the transition to that converged Ethernet network. With the stacked VLAN (802.1ad) and provider bridging (802.1ah) enhancements to the Ethernet standard, the customer separation and management infrastructure is emerging that will allow carriers and their customers to share a common Ethernet domain. Concurrently, the Resilient Packet Ring (RPR, or 802.17) standard provides for simple transmission of Ethernet frames in a ring-based topology and enables economical telecom-quality reliability. Both of these developments are symptoms of a dominant preference for packet-based over circuit-based networks.
The widespread adoption of Voice Over IP (VOIP) technology both throughout the enterprise and increasingly among the carriers themselves is helping shape this preference. By defining Quality of Service (QoS) requirements and taking advantage of the bursty nature of Ethernet traffic, VOIP allows enterprises to merge their voice traffic onto the data network and dispense with separate PBX equipment. The Power Over Ethernet (802.3af) standard was an important milestone in this effort, allowing a copper 10/100 Ethernet connection to supply power to the phone unit and successfully replicating the existing telephone service model.
VOIP technologies are also quickly transforming the residential voice landscape, as customers choose to receive voice service over either their cable/DSL connections via a third-party provider like Skype or Vonage, or directly from a cable company competing against the incumbent phone monopoly. Even the traditional voice providers themselves are moving in this direction, with British Telecom perhaps the most visible proponent--they intend to offer only a fully converged packet network by 2010.
What to do?
So where does this leave the networking equipment industry? Because we are still in the midst of this transition to a packet-based, largely Ethernet world, the old means of providing voice and data services cannot be ignored or marginalized. The installed base of SONET transport equipment, particularly in North America, exceeds several billions of dollars and still has many years of useful service ahead of it. ATM is still deeply entrenched in both the wireless and access markets. And, the migration to an Ethernet metro service model is steady but slow, leaving most businesses still relying on T-1 connections for Internet and VPN connectivity.
At its root, today's convergence means offering a single packet network while preserving the flexibility to integrate legacy connections. This can take a variety of forms: multiple protocols on an individual port, split circuit and packet capabilities, or enhanced QoS. Let's consider each of these in turn.
The core transport SONET network has already converged on MPLS for packet routing in most large providers. But even within this 'unified' network there are multiple widely-installed Layer 1 and 2 protocol choices that must be supported, including Packet Over SONET, ATM, and Ethernet Over SONET. Because these are mapping choices, they can be efficiently supported in the interface silicon of the carrier equipment. With today's highly integrated silicon it is possible to support them all simultaneously on the same IC, reducing the number of discrete line cards that the system vendor must offer.
An additional requirement is supporting the multiple rates of SONET, which range from OC-3 (~155 Mbps) to OC-192 (10 Gbps). Again relying on the power of advanced semiconductor integration, many of these can be integrated on the same device and provide an additional lever of flexibility to the system vendor. This integration extends all the way to the optical module as the interface silicon can include a multi-rate serdes as the line interface, with all framing and mapping work completed in the device.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.