Brightlink Gets Second Chance

Nearly two years after Brightlink Networks shut its doors, its technology is in the hands of two former employees with dreams of rekindling the company's radical approach to large-scale switching.

Having picked up Brightlink's intellectual property on the cheap, the pair is hoping to license the technology. They've gotten nibbles related to large crossconnects and STS1 grooming switches -- the kinds of work Brightlink's BOSS 1000 would have performed. But Kerry Davis, who founded Aztech Partners as a holding company for Brightlink's technology, wants more.

Davis thinks Brightlink could have radically changed the way large switching platforms are built -- if only the startup had realized its own potential. His goal is to find a licensee that will be willing to build products based on a distributed switch fabric.

"The technology was never presented this way. The marketing and sales groups at Brightlink were not of a TDM background," Davis says.

Brightlink's run was short but eventful. The company started in 1998 as Uniant, a chip firm, then changed its business model to systems and its name to Corvia. Lawsuit threats from Corvis Corp. (Nasdaq: CORV) prompted one more name change, to Brightlink. The startup was building a crossconnect that could scale to large port counts -- a big requirement at the time -- while still shuffling traffic in STS1 increments.

Davis managed Brightlink's Sonet system architecture team and opened the company's Texas facility, but he departed amid disagreements with management's direction. As Brightlink collapsed, Davis worked a stint at Polaris Networks and later pursued his own ideas. "I had at the time talked to some VCs about doing something with a VT1.5 type of switch, which would be an adjunct to the STS grooming switch that Brightlink had," he says.

Meanwhile, Brightlink's intellectual property -- much of which hadn't yet been patented -- was still on the block, with no takers.

"Once the price got down to a reasonable range, I just pulled in a couple of partners and decided to finish the patent work," he says. (His other partners, who prefer to stay anonymous, are a fellow Brightlink alumnus and an associate experienced with Telcordia processes.) "Up until around Christmas [2003] we were just sitting on it. Then we started getting calls."

Calls? Yes, word got out about the bankruptcy sale, and some interested parties tracked their way to Davis. The volume of calls prompted Aztech to create its Website, which got 3,700 hits in its first week, Davis says.

Beyond email sent to some prospects, Aztech hasn't publicized its efforts much. "We're just hoping to get the technology back out there," Davis says.

Quite a few telecom companies have been the subject of management buyouts (see Management Buyouts Blossom, Germany's Aifotec Is Reborn, Axsun Gets ISO 9001 Certified, and Billing Online UK Signs Up 2). Aztech's licensing approach carries less risk than trying to rebuild a whole company, but revival of any technology is tough, especially when customers have already passed on the idea. "There aren't many successful examples of licensing the intellectual property of a company that's been foreclosed on," says Susan Mason, partner with Onset Ventures.

The technology's degree of eventual success might depend on how involved Davis and his Brightlink partner get.

"It’s not outside the realm of possibility that [a technology-licensing startup] could work, if you can also capture the critical talent to accompany the intellectual property. If not, then it’s largely worthless," says Tim Danford, technologist-in-residence with Storm Ventures. Scaling up
Davis's ambitions revolve around the idea of a distributed switch fabric. Most large routers and switches use a centralized model, where the switch fabric -- the matrix that connects one port to another -- sits in a single physical place, often on one card in one box. All other cards inside the box must connect to it, and, in multichassis implementations, every other box must connect to the switching box.

This leads to a problem in scaling. As the system grows, more of these connections build up, and the architecture can't keep pace. Besides becoming physically entangled, the switch might encounter problems balancing all the traffic flows properly.

One possible answer is to split the work among the different linecards and boxes, distributing the switch fabric's workload. The result is a simpler and cheaper architecture, one that uses fewer chips to get to higher port counts. That's what the BOSS 1000 did -- by placing a switching ASIC on each linecard, it didn't need a switch card.

Router vendors Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7) and Hyperchip Inc. are using distributed fabrics but don't have huge success to show for it -- although, as Davis points out, they're targeting the IP routing market, whereas Brightlink was aiming for Sonet transport.

To Davis, the problem lies in cracking the mold of tradition. The Clos architecture, which dictates the "normal" layout of a switch fabric, has been around for 50 years. It works, and systems designers are familiar with it.

Davis thinks his best bet might be to get a chip company interested, because few equipment vendors can afford to take chances on something new. "Because money is so tight, the people who are building equipment are really short-sighted," Davis says. "If you had a company that did for switching platforms what the PMCs and AMCCs did for Sonet chips, then all these startups could come out doing I/O shelves" instead of systems -- a less expensive proposition that wouldn't force startups to compete with switches from Alcatel SA (NYSE: ALA; Paris: CGEP:PA) and Tellabs Inc. (Nasdaq: TLAB; Frankfurt: BTLA).

Another idea, which Brightlink never explored, is an alternative to the God Box. Rather than create one system that handles all traffic types (ATM, Ethernet, Sonet/SDH, etc.), it should be possible to keep separate systems but use a BOSS-like fabric to interconnect them. "You could have an ADM which shares a switch fabric with a crossconnect which shares the same switch fabric with a router," Davis says.

Whether the Brightlink technology sinks or swims this time, Davis and his partners might have found another calling in Aztech. "We're working on other things that it's too early to advertise."

— Craig Matsumoto, Senior Editor, Light Reading
Page 1 / 3   >   >>
Ringed? 12/5/2012 | 2:20:17 AM
re: Brightlink Gets Second Chance Now that the fog has lifted Brightlink would build a DACS that provides narrowband and wideband grooming and switching.. Novel concept.

There is really only one major issue left to be addressed. FUNDING! Hopefully they are smart enough to realize that even with a intellectual property headstart it will still take north of $100 million to pull this off.

Besides. The last time I looked chip companies had taken the same beatings the equipment vendors did.

eyesright 12/5/2012 | 2:20:16 AM
re: Brightlink Gets Second Chance Say, you could probably use some ASICs from a certain vendor based in Conneticutt...

...I hear that their San Jose based QA department just about has all the bugs worked out.
materialgirl 12/5/2012 | 2:20:10 AM
re: Brightlink Gets Second Chance How much would it cost to tape out an ASIC? Would an FPGA work and do the job for less? At least to attract money?

NT has made it clear that they are becoming a software company. The telco industry is finally migrating to the more efficient PC model, where the hard work, say of making a CPU, is done once or twice, with the cost then amortized over the entire industry. It pays off big, both for users and for the industry. I think this is a good sign. As spending comes back, those chip guys will do just fine.
OpticOm 12/5/2012 | 2:20:07 AM
re: Brightlink Gets Second Chance Despite the fact that the market is still tough, and money not on every fence, they might stand a chance.
The next gen SONET boxes need to be developed.
And the hint about a MSPP box, also makes sense...
opticalwatcher 12/5/2012 | 2:20:04 AM
re: Brightlink Gets Second Chance Wasn't there an article a while back about the intellectual property of defunct Crecent Networks being bought by someone, who was going to try to resurrect it?
Ringed? 12/5/2012 | 2:19:54 AM
re: Brightlink Gets Second Chance ASIC costs are not fully measured until stamped out in mass from the foundry. That costs big bucks.

No. An FPGA will not work. They are too big and consume too much power for a high density high port speed design.

FPGA's have a purpose. WRT an ASIC they prove in the design if tested properly with assumptions.

You are right. Nortel is going to be a software company. And, Lucent will be a services company.

Sad but True.

Kerry Davis 12/5/2012 | 2:19:38 AM
re: Brightlink Gets Second Chance Wow! I just got back to the states late last night from a businees trip and didn't realize the article had run till this morning when I looked at my email from friends who had read it. 1600 hits to our website yesterday alone!

Some answers to some of those questions:

There are actually 3 of us in the partnership.

So that I don't slight the work of George Wheeler and some other very good people, I initiated the hiring and initial setting up of the Texas office. George and some other very fine people did much of the work in the day to day operations following that, while I maintained a small group of system architects there as well.

Yes, we would like to license out, but we are certainly not stuck on that as the only way to strike a deal. It depends greatly on the deal and the way the IP is to be used. For instance, there has been considerable interest from Asia to simply continue where BL ended and build an STS-1 DCS. A license would, in that case, not be in our best interest. Where it might work is in a generic switching fabric solution licensed to several TDM and PDU solutions providers who wish to be interoperable and save development costs without sacrificing the product uniqueness found in their different I/O solutions.

NO, this is not just applicable to TDM applications. Don't routers use SONET links and networks for transport? At the risk of being overly simplistic, just combine a 10G IP router solution chip set with four or sixteen of our distributed switching fabric ASICS on a one to four blades, bake in a pizza box, interconnect with east/west CWDM links and you have a 1-4 RU very scalable PDU/TDM solution.

But I believe the bigger advantage is that the switch fabric is way easier to manage than Clos and it is both distributive and scalable. It doesn't really care if the PHY is TDM or PDU. Unidirectional STS1 crossconnects can be made in less than one SONET frame time (125us) to drain a PDU buffer. So a Router from one company can share a single switching domain with an ADM built by a different company with only a single 125 uS combined propagation delay. Please read the article on HTvsClos.pdf on our web site for more info.

Were we forclosed on? I guess that depends on what was meant by that. BL simply ran out of money when the bubble burst and demand with it. BL itself, I am told, never filed bankruptsy and is technically still a Delaware corporation with no assets to this day. We even got the URL in the buy. I don't remember anyone saying the idea was a bad one. Just an untimely one. There were certain aspects that some people didn't like. Like the Vcsel interconnects coming out of each line card. But that is a fairly easy thing to fix today. So I don't believe the concept was ever judged to be bad. But either way, revival of a technology that never actually made it to the field is very tough. And part of the reason is the misconception that it failed because it didn't work. Especially with a product as complex as this. Where the intellectual property is often judged right alongside the application and market it was first placed into.

To answer some of the posted comments: I think $100M from here is ridiculously high at this point. Thats not to say that one couldn't spend that much. But you would have to spend alot of money outside of Engineering to do it.

No need for an FPGA. The existing 2.5G ASIC works very well right now. Yes, I have some ideas on improved features, but the existing chip does a very good job of scaling to 2.4 Tbps as is, and it can be production ready very quick. Of course, cost reductions and a 10G solution are always attractive for the future.

Anyone wishing to contact me directly can do so at [email protected]

And much thanks to all the well wishers!
Peter Heywood 12/5/2012 | 2:19:29 AM
re: Brightlink Gets Second Chance Before Brightlink went under, the VP of marketing was telling me that the Brightlink switch had undergone tests at a carrier and done exceptionally well.

I recall him encouraging/imploring us to test grooming switches so that he could get some results out into the public domain.

We tried to do that, with BTexact, as some of you will remember. But all vendors wimped out on us and I suppose Brightlink had died by then.

In any case, I was wondering whether someone out there could now tell us more about who was tested the Brightlink switch -- I have vague memories of suspecting it was Williams -- and what the results were. I seem to recall the Brightlink VP claiming that other switches couldn't do some types of grooming.
OpticOm 12/5/2012 | 2:19:14 AM
re: Brightlink Gets Second Chance I think the unit is still passing traffic in a lab today at that carrier.
It proved to be a good product.
I will let the right gent to name the company...
unet 12/5/2012 | 2:18:57 AM
re: Brightlink Gets Second Chance You seem to be persistent in seeking information on BL both on this board and on the Switch-Fabric Chipsets article talk board.

any particular reason ? Probably if you ask more directly you may get some response.
Page 1 / 3   >   >>
Sign In