An Interview with Geoff Bennett, Director of Solutions and Technology, Infinera

Editor’s Note: At SubCableWorld, we often have stated that it will take a dramatic change in the forces driving the market for the current boom market to end.  This is particularly true in terms of bandwidth demand, which continues to increase at mind-boggling rates.

We recently had the pleasure to speak with Geoff Bennett, Director of Solutions and Technology at Infinera about how demand is driving the market and how technology is working to keep up with it. 

Geoff Bennett

Mr. Bennett: If you look at what happened in the submarine fiber optics industry just after the turn of the century, you can see that it was a supply-driven market.  You had all this capacity coming online and very little demand for it.  So what happened?  The market crashed and many of those cables were later purchased for pennies on the dollar. 

It’s funny how markets rarely learn from mistakes.  You would think that might be a little bit fresher in the memory.  I think we’ve actually been really lucky in the telecoms market.  Since that time, as far as I can tell, we’ve been in a demand-driven market.  Demand-driven markets are far healthier and sustainable than supply-driven markets.  Right now in the subsea market, you look at where the demand is coming from and the answer is from two companies, basically, Facebook and Google.  If you look at Telegeography data on used international bandwidth, over 85% is from the ICPs – Facebook, Google, Amazon, Microsoft and Apple.  But when you break it down a little bit further, Google and Facebook vastly take more than the other three.  That’s because of their advertising models. 

Google and Facebook talk about these things at conferences, so it’s not secret.  They can calculate how much advertising revenue they lose for each additional millisecond of latency that the user experiences.  So if you are on your Facebook page and you are clicking or swiping through various parts of the page, if you are not presented with an advertisement within a certain amount of time you’re going to move on.  And apparently, I guess they have to prove to the advertisers that the eyeballs have actually seen the advertisements.  If you lack the capacity to supply those advertisements, you’re not going to get paid. 

It also means that, particularly Facebook, is replicating information from different users in different parts of the world.  People have friends in Singapore.  People have friends in Australia.  So those friends will need the same kind of speediness of updates, so your information gets replicated all over the place.  It’s a bandwidth-wasteful model but the amount that it costs Facebook and Google in bandwidth is dwarfed by the advertising revenue they get because of the higher performance. 

So that’s where the demand is coming from at the moment.  There doesn’t seem to be any obvious way to be more bandwidth efficient. 

The funny thing is, there are many communications service providers (CSPs) that are establishing subsea capacity with the goal of selling it to other service providers, companies, etc.  When you look at the share of capacity for ICPs, it’s dwarfing the CSP demands.  But if you monitor the growth of that CSP demand, it’s growing like a weed.  It’s growing dramatically.  It’s just that the growth from ICPs is just so much bigger.  If you were to subtract the ICPs from it, you’d still have a really healthy market that’s growing.  It’s just that we have established this big component from the ICPs. 

So the demand side is healthy.  The ICPs are generating the volume of bandwidth but the regular communications service providers are still growing at a very, very healthy rate.  Their demand is growing and they’re the ones that are going to provide connectivity in a more capillary way.  These are the guys that will bring the connectivity out to where it’s needed as opposed to just connecting two mega data centers. 

On the supply side of things, there are a number of interesting questions that I think we can ask.  There are over 450 cables worldwide.  I would imagine that most of them are what we call dispersion managed cables.  These are cables that were designed before coherent technology came along.  You have alternating positive and negative dispersion fiber in the cable.  When you put coherent transponders on them you’re going to get massive initial bandwidth.  When coherent subsea transponders were first introduced in 2009-2010, they multiplied cable capacity by a factor of four and then a factor of 10.  That was fantastic.  You don’t get that very often. 

The question arises, are they the best cables to run coherent on?  The answer is no.  Now that you have the benefit of hindsight in understanding what a coherent transponder looks like and what its properties are and once you’ve figured out that you can compensate for chromatic dispersion in a coherent system, what’s the next barrier?  Well that’s non-linear effects.  And it turns out that we can mitigate non-linear effects by having a cable with high chromatic dispersion.  So if you’re managing the level of dispersion in the cable, it’s never getting particularly high.  So cables like MAREA, AAE-1, BRUSA, Seabras-1, these are all positive dispersion fiber cables with large effective areas, because you want a large effective area in the fiber to minimize your non-linear penalty. 

I am amazed at the micro-engineering of the optical fiber manufacturers like Corning. They’re able to craft these incredible waveguides and it’s all glass at the end of the day.  That’s an immense engineering achievement, I think. 

So once we have these nice coherent optimized fibers, we can put the latest transponder technology on there and step by step we’ve found more and more techniques to squeeze out the last sort of drops of capacity from each of these fiber pairs.

In all of these cables, we’re getting close to the Shannon Limit for that cable because every fiber pair has their own Shannon Limit.  Modern transponders can get closer than anything can to the Shannon Limit.  If you look at the classic Shannon equation, there are only two terms in there.  There’s a bandwidth term and one based on the signal-to-noise ratio.  And that signal-to-noise ratio term is contained within a logarithm.  If you were to double your capability within the signal-to-noise ratio arena, like build a better transponder, unfortunately you don’t get double the capacity because you’re inside a logarithmic term.  Whereas if you can double the bandwidth that you’re using, you do double the capacity because you’re in the linear term. 

There was a point where somebody said, “Hey why don’t we double the bandwidth in the fiber pair, and we can do that lighting up the L-band?”  That is absolutely true if your goal is to double the fiber pair capacity, C+L is a great solution. 

The problem with a subsea cable is that you’re powering it from the ends of the cable.  You can only get a finite number of amplifiers into that cable and still power them.  So by putting a C-band and an L-band amplifier in the cable, you’ve got twice as many amplifiers and double the amount of spectrum that you’re amplifying.  What you’re doing is doubling the fiber pair capacity but you’ve made no change to the cable capacity.  It’s got the same capacity if you only build a C-band cable.  But what you have done is save thousands and thousands of kilometers of glass.  You’ve got half as many fiber pairs with each fiber pair double the capacity.  That’s why that cable made economic sense, because it’s a transpacific cable and you’re saving a lot of fiber. 

But you not getting more cable capacity.  The guys like Google and Facebook, they need more cable capacity across the Atlantic and Pacific between their data center locations.  That’s why instead of doing C+L, they’re doing Space Division Multiplexing (SDM).  That’s where you actually de-tune the amplifiers.  You don’t drive the amplifiers to their limits.  First of all, you’re only amplifying C-band.  You don’t have the amps as closely spaced as you would on MAREA where the amp spacing is 55 kilometers.  MAREA represents the pinnacle of an uncompensated cable.  Even on other uncompensated cables, they don’t have amps as closely spaced as that.  It’s why you can close the Atlantic with a 16 QAM modulation on that cable and have a very-high-per-fiber capacity.