(Updated 7th March to reflect feedback from Hyperchip.)

Privately-held, Montreal-based Hyperchip, which was formed in 1997 and currently has 190 employees, describes itself as a systems company building the world's first carrier-class IP system that scales to over a petabit (1000 Terabits or one million Gigabits of capacity) (See Note 1). Hyperchip's stated aim is to "cost-effectively scale the core of the Internet with the same reliability and simplicity inherent in legacy voice networks".

The company states as its main asset a patented switching fabric designed to facilitate advanced QoS management and deterministic system performance. Hyperchip's carrier-class PBR-1280 core IP system is vaunted as "more than a core router" and is said to extend in-service from a single to multi-chassis system supporting more than 65,000 ports while being viewed and operated as a single system. Hyperchip says that it is targeting large incumbent and competitive carriers.

In conversation with Remi J. Makhoul, Hyperchip's Vice President of Business Development, Optical Keyhole attempted to gain a better understanding of the company and its aims. Clearly, the company's view of the future is a strong and committed one - why else would Hyperchip be building an allegedly "massively scalable" router designed to handle 65,000 OC-192s as a single switch fabric, spanning multiple racks and delivering petabits of capacity? If the company achieves its stated aims, according to CIBC, "This would clearly be a massive step function improvement…the product would address both the router and route table scaling issue for a very long time."

So how near is Hyperchip to showing us what it's got? According to the company, the moment of truth is near: "We're in the final stage of commercialization", says Makhoul. "We've been in talks with several customers and we're ready to launch our product for deployment in the coming weeks - March, April time."

Clearly, the idea of launching a new core routing product at this time - though Hyperchip would perhaps dispute the definition - must be about as easy as selling sunray lamps in the Sahara. Makhoul simply describes the prospect as "not the easiest thing in the world". However, the company truly seems to believe that a lot of the decisions, if not deployments, in the core router space will be made in the next two or three quarters. In Makhoul's words, "That's going to give the carriers enough time to test any new products before deployment".

According to the company, and this is crucial to what Hyperchip is about, "There's going to be a point of inflexion where there will be a new core IP routing system that's going to go above and beyond what already exists."

Plain old routers, in Hyperchip's view, "Are really good if you're an enterprise or 'Mom 'n' Pop' ISP but if you really want to have a reliable network then you have to buy something that's way beyond that - a core IP system. We believe that the big carriers definitely understand that dynamic and the smart carriers are starting to get there. The more people understand the complexity of these large networks the more incentive there is to move away from this traditional router to more of a core IP system."

Hyperchip describes this dynamic, if it can be termed as such, as analogous with the progression of voice telephony from analogue to digital. "It's exactly the same thing that we're seeing happening over the next 18 to 24 months. It really depends on who gets it right - if the technology becomes available it's really going to change the way IP networks are built. If nobody gets it right we're going to see the same kind of blood letting that the entire IP industry is going through right now."

Also, the company is unwilling to disclose if it has any firm orders at this stage, an issue that Hyperchips says it will address in June.

On this point, Hyperchip simply states, "We have enough money in our coffers to get us through to mid next year and we have the backing of our investors." Also, the company adds, "We're based in Montreal - our cash burn is roughly half what it would be if we were based in Silicon Valley. In terms of facilities, what we pay in a year is what other start ups in the Valley are paying in a month. We aren't faced with the high churn rates of employees either, luring people with BMWs and so on."

Optical Keyhole couldn't help wondering how it must feel to be a small start-up about to launch a core routing product at the present time. Makhoul admits the climate could be better, adding, "I think to say otherwise would be absolutely ludicrous. Carriers are certainly worried about the viability of their suppliers today."

Hyperchip, like all telecoms suppliers, also has need to worry about the financial health of its customers. According to Makhoul, "We're concerned about the staying power of the carriers we potentially work with as it's a big investment for us. Depending on the carrier, it's a nine-month to eighteen-month sales cycle so we have to be concerned who we're working with and they have to be concerned about the viability of their suppliers." However, Hyperchip states that there are various vehicles, such as Canadian government guarantees, that help Canadian companies export their goods.

NB: Recent investment house rumours have suggested that Procket may be on the brink of signing a resale agreement with Lucent Technologies which, if carried through, would appear to place it firmly in contention backed by the credibility of what has been - at least before the recent problems that have seriously dimmed a once charismatic corporate persona - undoubtedly the strongest technical and commercial telecommunications brand in the world.

However, in Makhoul's words, "We're about 12 months ahead of them." Siemens, Nortel, Nokia have also all had projects in this space, although these have been discontinued.

Though there may indeed be a market out there for Hyperchip's PBR product, some are of the opinion that the company's offering represents overkill for the near future (two, three possibly four years) compared to clustering T640s or the potential Cisco product. Putting the 960 Gbit/s per rack PBR product in context, take the example of the London Internet Exchange (LINX), the largest Internet exchange point in Europe. According to the carrier, "One sees around 30 Gbit/s peak flow, doubling roughly once a year. I guess there's another 30 Gbit/s exchanged off the exchange, and perhaps three or four times that again being brought in via transit providers. So the entire UK Internet may be blowing 300 Gbit/s."

It is certainly the case, according to this executive, that "Really big networks are already struggling with existing GSR and M160 networks as they really do run out of interfaces just clustering together those devices. Hence the T640 and Cisco 'next big thing', which appear to quadruple Gbit/s per chassis, and scale to numbers of chassis without eating external facing slots - without routing between chassis".

      .....PoP 1.....               .....PoP 2.....

      *---[]---[]-----------------------[]---[]---*

where '*' are edge routers, and each [] is a core router".

"At present there are eight fully fledged forwarding engines along this route. The new Juniper and Cisco routers with special inter chassis connections reduce that number to four (by eliminating the back to back cards in each PoP). The logical last step would be to chop that down to two."

However, it is the vaunted 'massive scalability' that is the Hyperchip design's main claim for attention. And while, as a scalable collection of chassis, it scales on paper beyond what Juniper (and possibly Cisco) have (or, in Cisco's case, may soon have), the need for that degree of scaling is somewhat doubtful.

In the end, price can always be a balancing factor but only really where the highest performance standards have already been reached; global telcos are not in the business of taking risks with their core nervous systems. However in the words of Optical Keyhole's source, "Of course, if it's half the cost of the competition (equivalent performance implied!) then one would look a bit harder…"

It certainly seems selfevident that this substantial barrier to new entrants also rises with the number of companies already serving the market, and that if Hyperchip wants to be No. 4, it is going to have to work much harder than the three who have already made it.

It may also be worth quoting another highly relevant viewpoint from Avici's Director of Marketing, Esmeralda Swartz, who notes signs of collapse and loss of momentum amongst aspirant competitors in the core router business, arguing that the four main new startups of the last few years (Caspian, Charlotte's Web, Hyperchip and Procket) have made no significant announcement about their technologies or commercial engagements with customers for between nine months (in the case of Hyperchip) and almost 15 months (in the case of Procket). Swartz regards this as clear evidence of further impending rationalization of the supply structure in the face of reduced market growth and telco difficulties.

Note 1: Hyperchip's switch fabric has the capacity to scale to over 65,000 customer facing ports, and while that equates to a petabit, Hyperchip has modified the statement 'The Petabit Routing Company' to the 'Core IP Systems Company' "as times have changed over the last couple of years". Hyperchip also notes that "scalability is very important to us as it guarantees longevity of service, but equally important is the reliability, resiliency and flexibility of our system".

Note 2: Actual capacity is 320 Gbit/s per chassis, full duplex, or 640 Gbit/s per rack.