Make Way for Cisco's HFR

Tuesday is the big day.

That's when analysts expect Cisco Systems Inc. (Nasdaq: CSCO) to announce the oft-delayed Heavy Fast Router* (HFR), the next-generation core router and successor to the Gigabit Switch Router (GSR) 12000 series.

It's been two years since Juniper Networks Inc. (Nasdaq: JNPR) announced its T640 next-generation router, and in the interim, Cisco has preached the 12000 series as a sufficient choice for new core networks, refusing to acknowledge rumors of HFR being developed in the background.

Meanwhile, competing core routers have emerged from companies including Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7), Chiaro Networks Inc., Hyperchip Inc., and Procket Networks Inc.

Quite a few HFR details have leaked out in recent months. From what our sources say, the system is a 16-slot, full-rack chassis with 640 Gbit/s of capacity. Installations of up to 18 chassis are possible, for a maximum capacity of 11.5 Tbit/s. Cisco is expected to take the leap of running a new operating system on HFR, replacing the company's Internetwork Operating System (IOS), which, despite its success in carrier networks, is often criticized as insufficiently robust (see Source: Cisco's HFR Tips the Scales and Cisco's HFR Gets Mod).But what else is under the hood? On the eve of the HFR's unveiling, here's what analysts and competitors are saying about the system.

It's built for speed

Sources say the HFR will be the first Cisco platform to support 40 Gbit/s per slot, and it almost certainly will support OC768c interfaces, possibly in its first release. The "c," for "concatenated" indicates one flow of 40-Gbit/s data, as opposed to, say, four OC192 feeds.

Cisco is slinging the big guns first, trotting out OC768c, quad OC192, and quad 10-Gbit/s Ethernet interfaces. Sources expect the slower interfaces, such as single 10-Gbit/s Ethernet, to come in later releases.

Physically, the HFR will be slimming down. The first release is expected to fit a 23-inch-wide rack, but sources say later versions -- including the half-sized HFR -- will fit the more common 19-inch rack (see Sources: Cisco Building 'Son of HFR').It's a family affair

The half-sized "son of HFR" isn't the only follow-up announcement planned. Some of HFR's software features will be latecomers as well, meaning the new software will lag IOS for more than a year, according to some sources.

The family of follow-up features appear to fall into two groups, and one source pegs their arrivals at 10 and 16 months after the initial HFR launch. The 10-month group includes GMPLS, IPv6 multicast, and VLAN support. The 16-month group includes both Layer 2 and Layer 3 VPN support.

It's debatable whether a core router needs full VPN functionality, considering most of the work happens at the edge. Still, a lack of VPN support is eye-catching, with VPNs being such a hot topic in routing, and competitors count it as a strike against the HFR. "Certainly in the core, you're looking the router to transport VPN. It has to be able to recognize the MPLS protocol," says Hema Ganapathy, senior solutions marketing manager for Juniper Networks.

It's got a new lookCisco wouldn't dare ditch the command-line interface (CLI) associated with IOS, because it's become so pervasive. But it does seem time to upgrade. Sources say the HFR is outfitted with the new Craft Web Interface (CWI), a browser-based, graphical interface, which should come in handy for managing large, complex HFR installations. For those who prefer it, CLI will still be available.

The switch fabric story

It's a complex topic, but the switch fabric is the heart of a router's design, and in this case it demonstrates the differences in philosophy between the HFR and other core-routing platforms.

The HFR's switch fabric uses what's called a Clos architecture, a classic model developed by Bell Labs researcher Charles Clos in 1953. The idea is to arrange switching elements in multiple stages -- three, in the HFR's case -- which saves on the number of links required to connect every input to every output.

The single- and dual-chassis HFRs will implement stages 1, 2, and 3 on one switch-fabric card. That means the switch card looks normal, ignoring the chips on board, and the operator wouldn't even have to know about this Clos thing. Where things get tricky is when multiple chassis are involved.In that case, the HFRs are all connected to one or two switching hubs, a setup that sounds analogous to the TX router that's going to interconnect Juniper's T640s (see Juniper Goes Terabit With the T640). These hubs contain almost nothing but Stage 2 switch-fabric cards, sources say. The satellite HFRs are then populated with a different switch-fabric card containing the Stage 1 and 3 switching elements. Thus, all traffic goes from one HFR (Stage 1) to one of the hub routers (Stage 2), then to the destination HFR (Stage 3).

This requires some planning. "You have to invest in enough '2' cards up front because the '1' and '3' cards must connect to this component. Basically, you can't 'pay as you grow' like you can with a modular fabric," says Esmeralda Swartz, Avici vice president of marketing.

By contrast, Avici's TSR system distributes the switch fabric among all the linecards. As more cards are added, the switch fabric grows accordingly.

But wait, there's more

Plenty of questions about the HFR remain. Among the most closely watched aspects of Cisco's announcement will be the reliability features, which could determine the HFR's suitability for carrier networks. The HFR's power consumption will be scrutinized, too, as competitors say the system is more power-hungry than any other core-router chassis.Some of the HFR's best aspects might not be in the numbers, though. At least one analyst says the system includes usability features that reflect a real concern for telecom operators' needs. "They've done a lot of things that carriers will just love," says Debra Mielke, principal with Treillage Network Strategies Inc.

We'll have more on that Tuesday.

Craig Matsumoto, Senior Editor, Light Reading