Designing at the network edge

Designing at the network edge
A funny thing happened on the way to the high speed multigigabit/second future of the Internet. A combination of factors — the worst economy in a decade, the disaster of Sept 11th, and an over-building of the network core -- has led to a severe slowdown in construction of the so called "information superhighway." As a result, the suppliers of the systems and devices are looking for new markets. The good news: they have found them at the network edge. The push is on for building the on-ramps and off-ramps and widening and improving the complex of routes by which traffic is funneled onto and off the high speed network core.

And, at the network core conditions are slowly improving and companies are beginning to make headway — not on the pure performance of their designs — but on their cost-effectiveness. According to John Bednarek, director of business development, C-Port Network Processor Group, Motorola (Austin, Texas), the past two years have been challenging for the network switch market. "Switch sales have dropped 23% from 2000 to 2002 due to service provider financial woes at the network core, where earlier service provider networks were built out to accommodate the insatiable future demand of bandwidth," he said.

Compared to the almost glacial market movement at the network core, the frenzy of activity at the edge has embedded network processor, router/switch vendors and service providers excited, and feeling the rush of expectations and opportunities as they move to address the new set of challenges emerging there.

Despite the fact that bandwidth requirements for now at the edge are quite modest -- from 10 and hundreds of megabits to one gigabit per second -- everything else about the environment is much more complicated. For example, there are a variety of physical network specifications to be considered, depending on the provider and the requirements of the end user. These include two or three different types of wireless networking specifications, Ethernet, ATM, cable modem and a variety of message protocols below the TCP/IP layer and atop it, varying security and reliability needs and wants. And, the media that users want to run over these various channels also varies — data only, data and voice, voice over IP, and multimedia — as does the quality of service levels.

"The multi-service edge network is rife with opportunity," said Rubin Dhillon, vice president, business development, SBS Technologies, Inc. "but hard work is needed to take advantage of it. For one, communications equipment manufacturers may have trouble finding embedded systems developers capable of meeting all of their requirements."

For all the opportunities the network edge environment offers, said Keith Morris, director of product strategy, at the switching and network processing division at Applied Micro Circuits Corp. (AMCC, Sunnyvale, Calif.) , it is an increasingly "messy" environment. "On one hand, the on-going proliferation of a variety of new services is creating significant opportunities for service providers to expand their offerings and to improve productivity for their subscribers," he pointed out. "On the other hand, tight capital expenditure (CapEx) budgets mean that service providers need to squeeze maximum utility out of every dollar invested."

Contributors to this week's In Focus concentrate on the design challenges at the network edge. Surmounting them will require more than just the imaginative use of existing network processor architectures, says contributor Vinoj Kumar, product architect at Agere Systems Inc. (Allentown, Pa.). Instead, fundamental modifications are needed that allow network service providers to offer nuanced, carefully differentiated and layered services.

A major focus area for system designers will have to be the creation of multi-service products that can offer flexibility for simultaneously handling traditional subscriber services, according to director of product marketing, Pradeep Shenoy at Teradiant Networks, Inc., (San Jose, Calif.). These include ATM and Frame Relay, along with packet-based next-generation services, such as IP and MPLS. The same platforms also need to be able to handle a wide range of interface speeds, from T-1 to OC-48.

Timing,cost issues

"Not only are there the issues of software programmable versus application specific hardware, but the important issue of time to market as well," said Shenoy. "The ASIC approach gives you performance, and the programmable approach gives you flexibility, if you have a sufficiently large software team with experience in all of the right tools and the protocols that need to be implemented."

Shenoy believes a better approach is a hardware specific configurable architecture that addresses a broad range of particular protocols, such as MPLS, but does not involve the huge investments in software and training that traditional software programmable approaches require.

"Although the public networks are converging towards IP/MPLS, there is a tremendous amount of legacy equipment deployed in existing networks that are generating the bulk of the revenues today," he said. "Multi-service switches have to support the legacy protocols and their transition towards IP/MPLS." As edge systems, multi-service switches also need to provide tunneling functions, flexible classification, and bandwidth management on a per-customer basis — a significant investment in software support.

While keeping costs in mind, the systems designer and the software and hardware vendors face an array of technical challenges that are equal of, if not more complex than, those at the network core, according to Bruce Hunter, network processor marketing manager at the Motorola Computer Group in Tempe, Ariz. "For one thing, in addition to being able to handle multiple protocols in the same box, and to interwork between them, there is also the need a need to handle both voice and data as well as to provide value added services, such as IPSec and content-aware switching," he said.

And because the network environment is still in a stage of flux, there is the need to update equipment as new protocols are introduced and to change the mix of protocols and services in a box. "3G wireless networks need to be able to migrate from all ATM to mixed IP-ATM to all IP while simultaneously increasing the ratio of data to voice bandwidth," said Hunter. "Similarly, ATM boxes will need to migrate to MPLS and IPv4 boxes will need to migrate to MPLS and to IPv6." It is also necessary to scale architectures from a small number of users, such as may be required by SOHO applications or satellite offices, he said, to an enterprise-wide solution that may need to support thousands of users.

Agere's Kumar, pointed out that current service layer agreements (SLAs), impose some difficult requirements on the traffic management devices. "Premium services, for example, require true bandwidth and delay guarantees," said Kumar. "There is a need to shape a large number of flows, which translates to the ability to support a large number of queues. Routers must also support dynamic bandwidth management, offer rapid service provisioning, and support multiple protocols and services."

Also, rapid service provisioning capability means that the network processors support efficient traffic management functions and be highly programmable. "These network processing elements are expected to support per port flow control, and the ability to interface to external fabric schedulers — without compromising performance," said Kumar, "or cost constraints".

Just as cities and communities who have on-ramps to freeways and highways have to come up with procedures and rules to ease the traffic problems that result on the ground streets, traffic control and policing is also needed in network core and edge applications to smooth out traffic bursts, Kumar noted. "Congestion occurs when packets arrive faster than the output interface can send them out," he said, as a result, "priorities need to be set for packets, perhaps based on the SLA or application, and either be dropped or queued for transmission." He noted that queue scheduling and shaping functions are closely related to congestion management. Because a switch/router is a shared resource in a network, many applications and users compete for the switch/router's limited resources such as memory and port bandwidth. A queue scheduling discipline allows management of these resources while guaranteeing service level agreements.

Realistic options

While networking companies such as Agere are now introducing single chip solutions that integrate such functions, it is not at all clear to some, like contributor, Greg Christison, communications processor development manager at Texas Instruments, that there is an appropriate alternative at the network edge, in particular. "Broadband applications are still evolving, and solutions providers require time for the design and integration of modules for new peripheral interfaces," he said. "And sometimes the market simply does not justify integrating a specific interface into a broadband processor."

In these cases, he noted, off-chip modularity makes it easy for the system developer to add a peripheral interface to a net edge customer premise equipment product with minimal hardware and software development beyond that required for the functionality of the interface itself.

"An easy-to-use modularity for external peripherals interfaces allows the developer to avoid risk by scaling up to a large design through successive smaller designs," said Christison, "rather than by undertaking the large design all at once. For the processor itself, retaining some interfaces as off-chip options saves die space and reduces pin counts, with a corresponding savings in cost." Because CPE manufacturers must provide products to meet a variety of broadband applications, he said, it is important that present and future silicon solutions offer flexibility and cost-efficiency along with performance.

According to AMCC's Arvind Ahuja, a Network Processing Forum board member, while software programmable NPUs are lower cost than hardware specific approaches, vendors and service providers need to keep software development costs under control as well. "What many service providers want is some sort of transparent way to adapt software developed to address particular protocol needs at the core of the network to the needs at the much different and more diverse network edge," he said. "They also need a way to seamlessly translate that program development cost across different NPU platforms."

He said that several of the software expert groups that he is involved with at the Network Processing Forum are moving toward the final stages of a common set of Application Program Interface specifications for many of the protocols necessary that will allow this sort of seamless migration.

Ahuja said that the NPF is in the final stages of approval for of an API for the current IPv4 and is at work on the next generation Ipv6 specification and for MPLS. Forum expert groups are also looking at the need for another one focusing on the develpment of an API aimed at supporting high availability software operation in routers and switches.