Optical components

HP's Newer, Simpler Optical Backplane

PALO ALTO, Calif. -- Researchers at HP Inc. (NYSE: HPQ) are trying out an optical backplane architecture that appears simpler than the normal approach and could obviate the need for a switch fabric.

What makes it simpler is the use of a bus carrying traffic past every line card. The normal approach is to replace the electrical connections with optical ones, which leaves the same number of connections but makes them faster and more expensive.

"You're not doing anything new" with that approach, says Michael Tan, an HP Labs distinguished technologist. "One of the things one of our Fellows talked to us about was, if you want to replace copper [that way], you'd have to get plumber's wages."

The press got its first look at the backplane Tuesday at the HP building -- the one where Bill Hewlett and Dave Packard's offices are still preserved, as is the 1950s exterior. The occasion was an introduction to the HP Networking and Communications Lab, a key element in the company's newly found drive to chip away market share from Cisco Systems Inc. (Nasdaq: CSCO).

HP had already showed off the backplane at Interop in May, in private sessions.

The key materials are available now, but HP officials are saying the optical backplane is at least two years away from being a product, partly because they want to squash more cost out of the architecture.

"We believe we can be of equivalent cost, and perhaps 20 percent lower cost, than our electrical counterpart," Tan says.

A cheaper optical backplane could be a breakthrough. It's easy to presume that switches and routers will have to go to optical backplanes someday, when factors of speed, power and/or heat make copper traces infeasible. But the industry hasn't gotten there yet. Copper looks like it's going to run at least 25 Gbit/s per line -- suitable for dealing with 100 Gbit/s traffic -- while optical alternatives remain relatively expensive. To date, sticking with copper has been a no-brainer.

One word: plastic
The core of the backplane is a plastic spine that runs past all the line cards -- a stick with 12 waveguides running down it, essentially. Two of these got installed in the 8212, each carrying traffic transmitted by three of the box's six line cards.

The backplane itself is dumb; traffic is simply sent to every line card. A semitransparent mirror diverts some of the optical power to each line card, and a field in the packet header indicates what the packet's real destination is. The line card ignores any traffic that's not meant for it.

The light coming from each line card is a set of four 850nm beams, each carrying 6.25Gbit/s of traffic. (Bandwidth-per-beam is limited by the electronics on the line card.) The waveguides are wide -- about the thickness of a human hair -- so that dust won't be a problem, Tan says.

HP wants to refine a few things before trying to make the backplane into product. The electrical-to-optical interface isn't yet integrated into the line-card ASIC, for instance. And once the signal becomes optical, it's sent to the backplane connector via fiber; researchers want to replace that with on-card waveguides.

— Craig Matsumoto, West Coast Editor, Light Reading

Page 1 / 2   >   >>
alandal 12/5/2012 | 4:56:31 PM
re: HP's Newer, Simpler Optical Backplane

Yes? Then at least 2 years away from any real products, maybe more ...

sam masud 12/5/2012 | 4:56:30 PM
re: HP's Newer, Simpler Optical Backplane

Why 2 sticks? Is it because one carries traffic one way, and the other in the opposite direction? Just curious...

BigBro 12/5/2012 | 4:56:29 PM
re: HP's Newer, Simpler Optical Backplane

Interesting demo. Basically, they're running 4 lanes of RXAUI over a shared optical bus on SR optics.

The killer in the real world is going to be dust -- network gear is not typically deployed in cleanrooms.

Duh! 12/5/2012 | 4:56:29 PM
re: HP's Newer, Simpler Optical Backplane

I've occasionally seen schemes for shared optical physical media.  I can't recollect one that involved plastic waveguides with embedded half-mirrors, but haven't followed the space consistently, either.  Looks very interesting.  I'm curious  about how they couple the line cards to the waveguide, how they fabricate the waveguides and plan to achieve manufacturing yields, how they do bus access arbitration and so on.   In any event,  I'm looking forward to technical papers in JLT and/or OFC. 

Rush21120 12/5/2012 | 4:56:26 PM
re: HP's Newer, Simpler Optical Backplane

" factors of speed" is incorrect, any one who knows physics knows plastic and glass have much more latency then electrical conductors like gold, copper etc.   This should be stated as 'capacity' not speed.

This is an interesting concept but lacks manufacturability, cost would be expensive (vs. FR4) and useability (image customers shaking or dropping it).  Getting optical down to the cost of electrical backplanes will take some doing.  The other concern is additional opto-electronics will be required for modules, increasing price as well as compensation compontents (higher speeds, filtering, etc.).

While HP feel they have inovation, which they do, the practicality is unclear - can't wait to see how HP addresses it, it would be pretty cool having photonic backplane, PIC and photonics I/O; but I don't think the manufacturability for the masses are any where in the near term.

^Eagle^ 12/5/2012 | 4:56:26 PM
re: HP's Newer, Simpler Optical Backplane

HP presented this in detail at OIDA earlier this summer.  In that presentation, they gave us answers to all the items you brought up.

Rush, you are making several "assumptions" with no detailed knowledge of data on the HP approach.  

Based on what I saw, HP has thought of and has developed answers and solutions to all the items you bring up.

we all know what "assuming" means.


^Eagle^ 12/5/2012 | 4:56:25 PM
re: HP's Newer, Simpler Optical Backplane

You made several statements that you presented as fact instead of as your opinion.

1) not manufacturable

2) cost


Those were assumptions with no real knowledge.  I too prefer reality.

I say wait until HP decides to release more information.  

The presentation I saw looked pretty complete.  And I know the scientists and researchers on this project are quite good.  World class in fact.

So, I am willing to give them the benefit of the doubt and take a wait an see approach.

Since I do know a bit about photonics and also FR4, and since I saw that presenation, I think HP has figured out cost side issues.  



Rush21120 12/5/2012 | 4:56:25 PM
re: HP's Newer, Simpler Optical Backplane

Great then Speedster, when's GA? 

An I never assume anything, nore believe anything.   Reality is always better.

^Eagle^ 12/5/2012 | 4:56:25 PM
re: HP's Newer, Simpler Optical Backplane

I don't know if you are involved in OIDA at all.  But if you do, there is a presentation on the HP approach in one of their archives.  HP showed this solution at an OIDA meeting earlier in the summer.

Note: I never see Lightreading staff at OIDA meetings, nor at OIF meetings, nor at IEEE  meetings.   so often LR only gets the information that is included inside neccessarily short press releases. They might attend some of those forum's, but I have never seen them there.


Duh! 12/5/2012 | 4:56:25 PM
re: HP's Newer, Simpler Optical Backplane

'fraid I haven't been involved in OIDA for a while, and it looks like the technical stuff is password protected.

Page 1 / 2   >   >>
Sign In