EE Times - August 6, 2007 - (Page 8) News analysis <<1 ROUTERS else is today.” The center expects to release final results in November. If the system proves itself, the center could purchase some of the devices for use in two large Internet service centers it manages. Flow control and other concepts such as Layer-2 routing will become adjuncts to traditional packet routers, said Mambretti. “The general trend is moving from a one-size-fits-all approach to routing to a world where you have different techniques to handle undifferentiated and deterministic traffic,” he said. David Vorhaus, a research associate at the Yankee Group briefed by the startup, said, “The Anagran concept is one all service providers are interested in— injecting more intelligence into the network. The question is whether this is the best way to do it.” Vorhaus noted that major router makers such as Juniper Networks are working with companies that make standalone systems that plug into a network to handle some of the flow control work. “Most service providers already have lots of routers and they find it cheaper to www.eetimes.com and passed to the appropriate output port if it is part of a known flow. Only packets from a flow yet to be identified are passed on to a single merchant 5-Gbit/s router chip from Intel Corp. Conventional routers may use eight to 10 complex router ASICs handling data at up to 10 Gbits/s to handle each packet effectively. The router chip in the FR-1000 assigns a data rate and port to new flows as needed, based on knowledge of the flow’s needs and the capacity of the router. By contrast, packet routers put data into sometimes deep output queues that can create delays of about 0.1 second. In addition, traditional packet routers tend to give equal weight to data, voice, video and peer-to-peer traffic. This approach tends to give peer-to-peer traffic too much attention and video and voice too little, Roberts said. “You wind up with 5 percent of the people on services such as [peer-to-peer media site] BitTorrent using up 80 percent of the bandwidth,” he said. The FR-1000 observes the rate, duration, packet size and throughput of flows, adjusting dynamics as needed. This way, Cisco cofounder launches optical transport system merging from a long, selfimposed exile in Redmond, Wash., Cisco Systems Inc. cofounder Len Bosack this week will unveil the optical-transport fruits of his company, XKL LLC. Called DXM, the transport system is intended for enterprise and metropolitan-access networks, combining a dense wave-division-multiplexing architecture with simplified Internet Protocol mapping. It would not be accurate to call XKL a startup, since various forms of the company have been around since 1991, focusing primarily on I/O when it was known as XKL Systems Corp. XKL LLC has been reorganized since 2002, however, moving from an early focus on Layers 3 and 4 protocol development to optical transport for smaller-scale LAN and MAN networks. Bosack has kept a low profile since leaving Cisco 17 years ago. His cofounder and former wife, Sandy Lerner, has been in the public eye more often, founding Urban Decay Cosmetics and investing money in a restoration of Jane Austen properties. Bosack, meanwhile, has kept his head down and paid attention to networking trends in both the public data and data-center worlds, where server clustering and effective box-to-box I/O are more important than routing. In its early years, XKL was interested in handling I/O bottleneck problems for customers ranging from Digital Equipment Corp. to MCI. Integration is key However, since the optical boom at the end of the last decade and the telecom crash that opened this one, Bosack said he has paid close attention to the type of customer that can benefit from XKL’s transport work. “We don’t expect to sell a DXM to the likes of Verizon, and in the current state of the industry, we wouldn’t want to,” Bosack said. The problem in appealing to an enterprise audience is that the 1U DXM transport unit shows advantages in ease of management and in very fast (submicrosecond) path protection, but does not blaze radical new grounds. Bosack said the most important factor in designing what will eventually be a family of optical transport devices was to make sure they would integrate well with existing IP networking r4outers and switches. But Sterling Perrin, optical analyst at Heavy Reading, said that XKL still must show where its system can be differentiated from transport platforms from players such as Nortel Networks Inc. and Adva Optical Networking Inc. “They may be satisfied with an initial small percentage market share from Fortune 500 or 1,000 customers,” Perrin said. “But the expansion into a family of products should show more differentiation from what’s out there than we have seen so far.” Transport flexibility Robert Michaels, business development director at XKL, said that the ability to easily configure available ports into ten 3-Gbit/s ports, ten 10-Gbit ports, or six 3-Gbit and four 10-Gbit ports should provide a degree of transport flexibility not available in most single-box units. Hardware redundancy and perchannel submicrosecond path protection also should be unique, he predicted. Among products on the drawing board is a band combiner, to provide expanded services over 40 discrete DWDM channels. The DXM architecture can work with point-to-point, linear add-drop and ring topologies. Because of the system’s size, it does not use separate line cards. Multirate XFP and SFP optical ports are integrated directly on the motherboard. The system is transparent to Layer 2 and 3 protocols. In an existing trial with a Los Angeles ISP, the DXM system is carrying Ethernet, Sonet and time-division-multiplexed traffic with the same path-protection speeds for all environments. “This is designed to appeal to the network engineer who has been a little bit afraid of optical products in the enterprise,” Bosack said. “From a management perspective, this looks like a network component, not a transport component.” —Loring Wirbel layer in a new technology, so the price advantage of an integrated flow router may not be that compelling,” Vorhaus said. “Also, they may have problems as a startup convincing service providers their system is reliable enough.” Addressing that concern, Anagran will aim some of its efforts at up-and-coming service providers building new WiMax and Fibre-to-the-Home networks. “There are big problems in each of these areas,” said Roberts. All sides agreed that Anagran is taking a novel approach by thoroughly inspecting packet header information and integrating flow control into the heart of a router instead of in a separate appliance. “The big companies will have a hard time with this because these systems have much lower margin and they will have to redesign everything,” Roberts said. How it works The FR-1000 sports up to 48 Gbit Ethernet ports. It saves information on all recent flow traffic in a bank of about 400 Mbytes of RAM. As traffic enters the system, each packet header is read by a Xilinx Virtex FPGA the system lets users make smarter decisions about what resources to dedicate to particular flows. That’s an advantages of the system the company needs to articulate more loudly, said Mambretti. “It allows you to allocate a kind of virtual pipe and stream data without disrupting other flows. That’s difficult to do with existing systems,” he said. Anagran has raised $28 million in venture capital, money that can take it at least through the end of the year. The company may raise an additional round to fund its expansion and a broad rollout of the systems, which are to ship this month. “We didn’t need as much money as other router startups that developed a lot of ASICs,” said Roberts of his company, which employs about 75 people. The current system is expected to address a broad market of potential carriers, service providers and large corporations that want to exercise better control of traffic typically on the edge of citywide networks. “This is a $10 billion market,” Roberts said. The company is considering a lowerend version of the router it may ship in about a year. ■ Electronic Engineering Times | August 6, 2007 http://www.eetimes.com
For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.