The purpose of this interview with Joe Kennedy (CEO) was to find out whether Pluris is, in the face of an uncertain market, happy to shelter behind the shield of being a private company, whether its view of the market is as bullish as that of Avici, and finally how it perceives itself as different from Avici.

Pluris has a headcount of more than 200, and recently completed its Series D funding, in which it raised $101.5 million. That round brought total funding to $162 million. See Appendix 1 for the full list of investors. The present funding level is expected to support the company through the first quarter of 2002.

Pluris' management consists of:

• Joe Kennedy (CEO): Ex Bay Networks, Rapid City Communications, Hughes LAN Systems, Phoenix Technology, Ungermann-Bass

• Bülent Erbilgin (VP Engineering): Ex NetFrame, MasPar, HP

• Keith Higgins (VP Marketing): Ex Copper Mountain, Cisco, StrataCom, CompuServe

• Warren Roddy (SVP Sales): Ex ADC/Pairgain, Nokia/Diamond Lane, Cisco

• Diana Everett (VP Finance): Ex Icast, NetFrame, Silicon Graphics, Tandem

• Dave Cox (VP Operations): Ex Cisco.

Pluris recently outgrew its main facility in Cupertino and has expanded into a second site near-by. It will move into new headquarters early in the third quarter of 2001, consisting of two new buildings currently under construction. Pluris' first office outside the US was opened last autumn, in Frankfurt, Germany.

Pluris' Teraplex routers are aimed at any core Internet backbone provider, large ISP, incumbent local exchange carrier, submarine carrier and other service providers.

The Teraplex routers are scalable and designed for the IP core. According to Joe Kennedy "Single-box routers available today cannot scale to increase port density or higher-speed interfaces. When you run out of room in today's routers, you buy a second one, use expensive I/O ports to link them together, and then have multiple routers to manage. The Teraplex architecture is designed to interconnect multiple chassis, and it uses internal switch fabric capacity, and not external I/O capacity, to link them."

The Teraplex employs a fibre optic backplane and chassis interconnect. This feature enables service providers to distribute chassis as far as several hundred meters apart in a central office or PoP, and it supports generational upgrades of interface cards because the fibre is speed-independent, unlike copper busses on a backplane.

Key parameters of the Teraplex router:

• Supports OC-12, OC-48, and OC-192 interfaces

• Can be configured to include as many as 128 chassis in a single router

• Fibre optic interconnect allows service providers to distribute the system, all acting as a single router, across hundreds of meters

• No single point of failure in the system - each line card links to two fabric cards, each link has a redundant path, and each fabric card is interconnected with other fabric cards. Single failures, and many multiple failures, do not disrupt traffic forwarding on the router

• Demonstrated interoperability with Cisco equipment at OC12 and OC-48 and with Juniper equipment at OC12, OC-48 and OC-192 rates via in-house testing.

Pluris' intellectual property resides in both hardware and software. In hardware, Pluris has developed six ASICs which are critical to its ability to deliver line-rate performance across all interface cards, up to 10 Gbit/s OC-192, with full QoS. Joe Kennedy noted "We succeeded in being able to ship system prototypes to our development partners on first-spin ASICs, a significant accomplishment given the high gate counts."

The critical piece of software Pluris has created is the distributed protocol stack. Pluris has developed distributed implementations of BGP, OSPF, and IS-IS so that multiple CPUs can participate in the route calculation algorithms. This element of scalability is needed to complement the extensive hardware scalability of the router architecture.

Imported technology
"There are several hardware components that system companies acquire, such as transponders, CPUs, memory, framers, and other parts. We also started our protocol development based on third-party code from Phase 2, subsequently acquired by Nortel. Routing companies no longer start with the RFCs and write each line of routing code from scratch. However, it still takes significant time and talent to integrate that third-party code, and Phase 2 wouldn't recognize it in its current form."

Optical interconnection between boards
If edge emitting 1300 nm transmitters are used, the chassis can be one thousand feet to several thousand feet apart, depending on whether multimode or single mode fibre is deployed. Using 850 nm VCSELs with multimode fibre, the range is approximately one hundred meters. "Most people are interested in just a couple of hundred feet rather than large distances, but we know that sooner or later somebody will ask us to go longer."

Integration
The Teraplex architecture has three card types. Control cards and line cards are very similar in architecture - both are 10 Gbit/s cards but the control cards do not have any I/O. The third card, the fabric card, makes up the 'cloud' of switching fabric, the periphery of which supports connections to the line cards. "How you hook together the individual switch elements of that cloud is what gives it its characteristic for forwarding," said Mr. Kennedy. Arranging the fabric in a tree structure, for example, suits a broadcast stream.

Joe Kennedy went on to explain that Pluris' product could never have been built a few years ago, simply because the accumulated latency through multiple elements "in a cloud like that would have given you outrageous jitter specs. Latency through the fabric today is measured in a few microseconds, so even in worse case scenarios the jitter and latency specs are still pretty good compared with what could have been achieved in the past."

Key targets and achievements

• The Teraplex architecture, if kept in a hypercube model, can scale up to 128 chassis, though Pluris does not expect service providers to build systems that large. "We just wanted to take scalability off the table as a limiting factor," said Mr. Kennedy.

• The use of optics means that Pluris suffers a 9 percent penalty in the power usage over a pure copper implementation, but the flexibility advantages far outweigh the small power penalty.

• Software deployed since June 1999

• System alpha trials underway at Global Crossing and Deutsche Telekom.

The biggest differentiator Joe Kennedy sees between Avici and Pluris is that Avici's architecture is hard-wired into a copper backplane while Pluris uses optical busses to interconnect its boards; and not just boards in the same chassis but boards that link separate chassis. Therefore the Teraplex router chassis do not have to be co-located - which Pluris sees as giving service providers a huge degree of freedom in planning a PoP layout. This design enables service providers to grow a system without extensive space planning, and they can use separate facilities, such as different buildings and power sources, to augment the reliability of the router.

Joe Kennedy further explained that the interconnect scheme between the switching elements in both the Pluris and Avici products are similar in that both companies use switch capacity to link chassis. "In Avici's router, the switching elements can look front and back, left and right, up and down, and you hook those elements together to form a torus. In the Pluris architecture, the default configuration forms a hypercube, but the larger difference is that in our product the configuration is changeable. It can be optimised according to the needs of the customer."

Both Avici and Pluris appear to have a solution that is based on a random distribution of traffic; i.e. traffic coming in on any port is as equally likely to go out on any other port. But "traffic often isn't that perfectly random" Mr. Kennedy noted. He explained that in the case of say a video distribution network, which is a one-to-many application, one would simply tailor the switching fabric into a tree structure, for example.

A further advantage Pluris sees in its solution is the savings in weight since the Teraplex does not have a heavy backplane. In Joe Kennedy's view it remains to be seen whether Avici can successfully upgrade to the next generation without changing its backplane - he acknowledged that it can be done "but I've never seen it done with copper."

Joe Kennedy views Charlotte's Networks' router as a metro product because of its support of TDM. "It is a super version of an access mux, allowing you to add a routing function to the back end of any heterogeneous service, which may be a bit more than people like Alidian allow." "However, I don't know that they (Charlotte's Networks) would agree with that."

Pluris acknowledges that one of the innovative things that Charlotte's Networks has done is to have an MPG interface on every port, and Charlotte's Networks claims that, as protocols change, it too can simply change software. This is viewed by Joe Kennedy as "quite an accomplishment if they can do that at OC-192 line rates." Pluris, like other players in the business they know, uses ASICs on the front end.

Mr Kennedy said: "In some sense, therefore, we are similar to Avici - as we are the only two scalable solutions currently being talked about." Though he noted that cluster solutions might also emerge from competitors such as Cisco and (previously) IronBridge, and perhaps Juniper.

Joe Kennedy stated that he defined "cluster" as a wheel and spoke architecture where standard routers are on the periphery of the wheel and they use MPLS or some other tag-switching mechanism to efficiently send traffic to and from the central switching core. "With that kind of architecture, vendors will have to convince their customers that the central hub is not a single point of failure and that the links to the hub have sufficient bandwidth. This evolutionary approach most often makes sense when a vendor has a lot of existing product in the field and they are trying to offer an interim update."

As for the future, Pluris is of the opinion that "anybody, five years out, that is serious about this business, will have a scalable solution, including Juniper and Cisco, maybe Procket, Caspian and Hyperchip. Certainly Cisco thinks it owns this space, but there is no denying it is losing part of it to Juniper on a regular basis."

Price per port is in line with the competition and no premium is charged for the scalability, because as Joe Kennedy says "you can't get away with it." He noted that: "However, the optics don't come for free and we need to keep the same kind of margins as everyone else."

Pricing was described by Pluris as a stair-step; i.e. Pluris is more expensive at the front end. "If all you are ever going to need is four OC-12 ports and two OC-48 ports then quite frankly you would be better off buying an M40 or something like that." However, if a customer has a situation where he needs the equivalent of two single-chassis routers' worth of ports, that is the crossover point where Pluris would have similar pricing, and beyond that the price curves diverge.

According to Joe Kennedy "the bigger the system the more you save due to the nature of the interconnect. The way that one interconnects standalone boxes is by using routed ports, so you are paying for routing, lookup, classification - the full cost of an I/O port plus the maintenance of two systems and other operational costs that go with a more complex, multi-box design. With the Teraplex, things are hooked together on the fabric side, using very simple crossbar switches. Economically, it is just far cheaper to do that than to use I/O."

As far as funding is concerned, Pluris has been much more conservative than many in terms of jumping into an IPO. In some sense it missed the peak of the market in terms of liquidity, but the goal was to keep the company intact. Pluris feels that it got a fair valuation on its last private round, and it has escaped the problems with having upside-down stock options now that the market is in turmoil.

Pluris faces the continuing challenges of a development stage company in its market; i.e. attracting and keeping the right people. Fundamentally, though Pluris is reliant on only the continued growth of the Internet - it is not dependent on holy grails such as broadband to the home or multicast.

Pluris also acknowledges that the industry is still struggling to develop some OC-192 parts, and that the parallel fibre components used for its router's backplane and interconnect are still emerging, though the landscape is rapidly changing because of the demand for parts to meet the VSR specification.

The main barrier to selling Pluris' product, in its view, is relatively simple to overcome. "We just have to convince the customer that a scalable, flexible solution saves him money, which frankly is pretty easy to do."

Pluris does not yet have committed customers, but does have two very strong development partners. Global Crossing has been trialing the system since September 2000, and has recently received a third round of equipment and code. Deutsche Telekom, which trialed the software last year on emulation platforms, began full system trials in February.

Pluris appears to be a well-led and confident company with a well-engineered product and a relatively clear view of the market. In addition it appears to be very well funded. Nevertheless Pluris is in a general market whose structure is not hugely propitious to the company.

Cisco, the dominant supplier is losing share rapidly but not particularly to Pluris' benefit, since the effect of Cisco's loss so far has been to effectively set up a competitor - Juniper Networks - which will soon be of comparable size to Cisco in the router field and probably considerably more dangerous. Meanwhile, it is not believable that Cisco itself will sink to the point where Juniper is larger than itself in routers without becoming fully competitive, so Pluris will therefore be faced by two pretty well equally strong, highly competent competitors, whose mutual competitiveness is likely to accelerate customer expectations and downward price trends. This is as opposed to the original market structure of a single dominant competitor (Cisco), who would have preferred to lose a small amount of market share rather than compete directly.

Meanwhile Avici is already out there and at best Pluris will be able to get only 50% of the attention of that part of the market interested in its proposition. As Joe Kennedy has indicated, several competitors are very likely to have a scalable product within five years and the two major companies almost certainly much sooner. As such the window of opportunity for Pluris to establish a position in this market is already narrow and until it starts to record revenues it will effectively get narrower as Avici, already with a product, builds position.

At the same time this is fundamentally a market with huge capacity for cost reduction through integration, as well as the ability to add lots more functionality, both of which are likely to demand steady substantial investment for some years.

There may also be a problem in that neither Avici nor Pluris believe they can get any price premium for scalability. (Presumably in fact there is a premium but the market discounts this against the fact that neither company has a commercial track record - i.e. without scalability both companies might have had to quote significantly lower than Cisco or Juniper for the same opportunity). This also poses the risk, if Cisco and Juniper do not see the situation the same way but regard scalability as a significant threat, that market pricing will become destabilised.

The good news for Pluris is that it has gained the interest of two of the world's largest operators. Provided it can convert interest there or in other large carriers into a commercial relationship and start to ship product it will be at least into Phase 3 of the company's development, and will have a toehold in the market which will hopefully be enough for the company to get to Phase 4 (which we define as achieving an inherently survivable share of say 10-15%). Bearing in mind that before Pluris are two much larger and competent competitors, one direct competitor slightly ahead, and a throng of newer competitors, some of which are proposing huge leaps in technology, Pluris is unlikely to have a particularly easy time of it.

Pluris investors:

• Lightspeed Venture Partners (formerly Weiss, Peck & Greer)

• Vanguard Venture Partners

• ComVentures

• Crescendo Ventures

• Bay Partners

• Chase Capital Partners

• Worldview Technology Partners

• Sands Bros

• ABN AMRO

• Alliance Capital

• Bank of America Securities

• Cassin Ventures

• CDB Web Tech

• Comdisco

• CSFB Private Equity

• Deutsche Bank

• Stephen Garofalo

• Global Crossing Ventures

• PB Ventures

• PCG Ventures

• Piper Jaffray Ventures

• Rana Investment Company

• RWI Group

• Samsung

• Special Opportunities Group

• Vulcan Ventures