From Network Briefing, a sister publication…

Tom Lyon, a 12-year veteran of Sun Microsystems Inc, and a founder member of the ATM Forum, was on a plane, en route to one of its meetings. He was thinking about the problems of integrating IP-based traffic with Asynchronous Transfer Mode networks, and he hit upon a brainwave – why not use high-performance hardware, but throw out the Forum’s software standards, and route IP natively. The idea resulted in the founding of Palo Alto, California-based Ipsilon Networks Inc. That was in 1994, and now an initially skeptical data communications industry is coming round to Lyon’s concept. Ipsilon says that it developed its approach, because of a fundamental conflict between existing switched local networks, and the applications they are designed to serve: while switched local networks are connection-oriented, most data networking protocols are designed to operate using connectionless technology based on global addressing. Although the industry has tried to get around this problem – in the case of Asynchronous Mode through offerings including Classical IP over Asynchronous Mode, local network emulation, and Multiprotocol over Asynchronous Mode – Ipsilon nevertheless feels that delays in bringing these to market, and the increased complexity that they add, mean that they are merely replacing one set of problems with another.

IP node

For its offering, Ipsilon has developed what it calls Generic Switch Management Protocol stack, which is designed to enable IP traffic to be routed across Asynchronous Mode hardware. The resulting system is claimed to provide a vast increase in speed, without a corresponding increase in price: the company’s flagship product, the IP Switch ATM 1600, is said to have a switching capacity of up to 5.3m IP packets per second, and port throughputs of 155Mbps, for less than $3,000 per port. This compares with the bridged approach of the Asynchronous Mode Forum’s local network emulation implementations, says Ipsilon, which have a theoretical maximum throughput of 2.8m packets per second. At its core, the IP switch implements the IP protocol stack on the ATM hardware; the hardware itself is designed to operate as a high-performance link-layer accelerator. A gateway provides the communications between the IP switch, and users’ existing networks. By implementing in full the IP protocol stack, the IP switch appears to the network as a normal IP node, preserving compatibility with IP applications, and network management systems. It can be run both in IP-only environments, and where other protocols (such as IPX, NetBIOS, SNA and DECnet) are employed through tunneling or encapsulation. Ipsilon says its system is fully compatible with existing IP protocols including Open Shortest Path First, the Routing Information Protocol, the Distance Vector Multicast Routing Protocol, and the Internet Group Management Protocol. Support for IPng, the next version of IP, the Resource Reservation Protocol, the Border Gateway Protocol, and the Intermediate System to Intermediate System Routing Protocol is planned. In terms of the hardware used, Ipsilon has taken Santa Clara, California- based MMC Networks Inc’s Asynchronous Mode switch chip set, which uses its ATMS 2000 system, itself based on MMC’s ViX architecture. The IP Switch ATM 1600, with 16 155Mbps ports, lists for $46,000 in the US.

By Matthew Woollacott

The IP Switch Gateway is designed to act as the access device for connecting existing local network and backbone environments with IP switch-based networks. Using Ipsilon’s Flow Management Protocol, it is said to provide IP routing capabilities and intelligent flow classification, and is configured through the company’s Network Voyager Web- based user interface. The gateway consists of Pentium-based hardware, which runs the FMP software, an Asynchronous Transfer Mode network interface board, and up to three PCI network interface boards – versions for 10Mbps Ethernet, Fast Ethernet and FDDI are available. Within its software, Ipsilon

has also implemented an intelligent classification scheme, designed to determine when traffic should be switched, and when it should be routed. Throughput of long- lasting flows – such as File Transfer Protocol, Telnet, HyperText Transmission Protocol and multimedia audio and video traffic – is maximised through the use of cut-through switching in the Asynchronous Mode hardware. Store-and- forward routing is used only for short-lived traffic, such as Domain Name Server queries, SMTP data, and SNMP traffic; Ipsilon estimates that up to 90% of traffic qualifies for cut- through switching. Whereas the original consensus of industry opinion was that the technology would fail – firstly because it is proprietary, and secondly because of the relative paucity of IP-only networks – one by one suppliers are adopting the technology. Initial criticisms of Ipsilon centered on two fronts: firstly, that the technology is proprietary, and secondly that it is applicable only to IP networks, while most enterprise networks are multiprotocol. Ipsilon has answers to both of these points. In fact, to counter the question that the technology is proprietary, it has published the Generic Switch Management Protocol on its Web page (ipsilon.com) as an informational Request for Comment: this is part of Ipsilon’s strategy to establish IP switches as a category in their own right, according to Blair. Indeed, one company – New York City-based Scorpio Communications Inc – has already converted the Ipsilon software to its Stinger 5 Asynchronous Transfer Mode switch and is shipping it for $3,000 per port.

Kalpana’s founders

Blair claims not be worried that Scorpio is taking advantage of the fact that the Generic Switch Management Protocol is now in the public domain, since Scorpio will still have to direct its customers to Ipsilon for its IP switch controller and its software, without which the system cannot operate. As to the question of the fact that it is IP-only, Blair says that we see IP as the strategic protocol. He also points to the fact that Kalpana Inc was initially criticized for being Ethernet- only; he himself was one of Kalpana’s founders. That notwithstanding, says Blair, the company is nevertheless exploring ways to switch other protocols – notably IPX – natively. Ipsilon’s approach is now winning over fans, the largest of which is Digital Equipment Corp. Last month, DEC said that, through a new strategic technical alliance, it plans to implement the Generic Switch Management Protocol on its GIGAswitch/ATM switch, and incorporate the Ipsilon FMP on its servers and local network switches. DEC says it will also work with Ipsilon on development of methods to handle traffic engineering, performance analysis, and routing hardware. Another advocate is Harris & Jeffries Inc, Dedham, Massachusetts, which just announced plans to implement the Protocol as part of its Soft-ATM line. Indeed, if the IP switch lives up to its early promise, it could be the rest of the industry – and not Ipsilon – that has to eat humble pie.