> Researchers say they have created fiber cables that can move data at 99.7 percent of the speed of light, all but eliminating the latency plaguing standard fiber technology.
If by "all but eliminating" they mean "reducing by 31%".
There's a hard limit to how low latency can go. We've been close to it for years. So a hop takes 100 ms instead of 150 ms? They're still in the same ballpark. To have a revolutionary impact, it would have to drop the latency an order of magnitude, which, as far as we understand our universe, is impossible.
I guess the real application is over short distances. I'd love to hear the ideal usecase that benefits from this improvement.
OK, say it does. But give it a few years for the technology to propagate, and all traders will have the same latency, eliminating any advantage it once had. So not really revolutionary in the big picture for that case.
Still not the case. The state-of-the-art networking technologies are expensive (300K+/yr per data center plus inter-facility connectivity running into the 1-5M/yr range)
Sure, Finance will be the first to adopt, but it actually has pretty big implications in larger internet use cases.
Lets say your application needs <50ms latency, low enough that your users can't notice. Today, you'll something around 10 datacenters in the USA/Canada to achieve this.
If cut that down by 31%, you might only need 6. (or less, this would take some real math to figure out).
But the impact is pretty large in the long run, if it can really cut long haul latencies by 31%.
The problem with this technology is that at a 3.5dB/KM loss, it's only really good for about 10KM. At such short distances, the latency difference would only be about 0.0167ms. Given all the other latencies involved between two end-to-end hosts, that's hardly worth laying new fiber for. Unless they manage to reduce the loss to within a reasonably small multiple of more traditional fiber, this new technology isn't going to have much adoption.
At only 37 channels, compared to the 80 that more traditional fiber can carry using DWDM, it also carries less than half the capacity. And that's assuming that it's capable of the same bandwidth capacity per channel; the article states 40Gbps per channel, whereas you can already go as high as 100Gbps over a single wave using DWDM, which means it may only be able to carry 18.5% as much.
Also, I don't see this technology being compatible with erbium-doped fiber amplifiers. See, even the best glass fiber is good only up to 80 km, that's not enough, but by placing a piece of erbium-doped fiber it acts as a laser-amplifier and the fiber lenght can go up to 4000 km. Now that's useful. IMHO the EDFA is the technology that really enabled high-speed internet.
Why couldn't these fibres work with EDFAs? As long as the fibres are designed to operate in the C-band (~1.55 micron wavelength), there shouldn't be a problem running them through an amplifier.
No physicist here but I think there may be losses in the interface air/glass. Also, air contains water and other elements that may present too much attenuation at 1550 nm. I suspect the different lambda is one of the reasons of the limited bandwidth of the hollow fiber. I should try to RTFA.
It's overcoming the losses in the interfacing that's the innovation here - holey fibres have been around for a while. Note that it's the same research lab doing this work as invented the EDFA in the first place.
Is the capacity limit of 37 channels a fundamental limit due to the design, or a consequence of fabricating these fibres in a university lab? I'm honestly asking since I can't access the full text at home, but I suspect it's the latter reason.
My guess is that the first customers for this cable will be in the finance industry. A 30% improvement in latency doesn't matter much for the layman, but in finance, it can mean significant amounts of money. Already, an enterprising trader constructed a more direct line from Chicago -> NYC [1]. Going from 16ms -> 13ms was apparently worth $300 million. This fiber could bring a very large profit if you had exclusive rights to it via a patent. You could execute guaranteed-to-profit arbitrage strategies between any two markets if you were willing to build out the network. And no one could compete with you for a whole 15 years!
- Can't failures be counteracted by having failsafe measures? For example, if the market goes down more than 5% in a few-minute period, the market temporarily shuts down for a cooloff period of several hours to diagnose what's happening.
- Isn't HFT irrelevant to long term investment, given that it erodes away short-term profit, leading to accurate pricing of the stock?
A failsafe/circuit breaker is a possibility but shutting down the market will affect confidence. I'd prefer to see a market operate on a tick basis I.e all trades for a tick are transacted in one go with a window between each tick and assertions before and after the tick processing.
HFT doesn't provide liquidity, it provides volume which is not the same thing as liquidity. Think churning the same 100 shares a 1000 times.
Since market participants assume that the volume they see is liquidity, when they try to execute larger orders, and BAM! (this happens with single large orders as well as a dump of 100 lots) the floor just vanishes from them as machines just turn off for some time while someone liquidates a position and tries to find ACTUAL buyers.
Wasn't this mostly stopped in NY when the Exchange set up a server building and allowed firms equal access? They had been buying up the real estate around the exchange for amounts much high than the property's value and it was becoming a bit of a problem. Now they are all in the same building and everyone has the exact same amount of cable running from their server to the exchange.
Changes in commodities prices in Chicago will have "known" affects on stock prices in NYC. If you can get the information between the two cities faster, then despite this equal access in the last leg you can still take advantage of knowing before the others.
This move halted the land rush around the fiber access points in New York. But suppose a firm trades both on the NYSE and CHX in Chicago. If there's a price difference on the same stock in the two exchanges, you can buy low and sell high for a brief period (a few ms): if your machines in the NYSE building and CHX building can communicate that price difference faster than anybody else, nobody knows there's a discrepancy to be taken advantage of.
As an aside, if this is rolled out widely, can we finally come to terms with the genius of Senator Ted Stevens metaphor about the internet being a "Series of tubes." :)
Half correct. Yes, people used to make fun of the comment all the time. But the only sentiment anyone has submitted on the topic in at least last 5 years is the exact thing you just said.
Financial field isn't the only one that cares about latency. I'm still waiting for it to become feasible to jam (play music together) remotely. This gives me hope, but the real solution is still far away. Interesting how despite all the tech progress this (deceivingly simple) problem is still unsolved.
It's off topic, and it's only as real time as the network allows, but this project is pretty cool and has a good community around it: http://www.ohmstudio.com
If you think about a recording studio experience, you usually lay down a rough basic track and overdub on top of it. In that case, latency doesn't really matter as much as good tools for collaboration and iteration.
no, there's lots of applications for low-latency connections. the reason algorithmic trading is special is that the financial people are the ones who have a valid business case for installing low-latency connections. is it worth a couple of billion dollars for you to jam remotely?
Few years ago the main business case for low latency was telephony. And that is the reason why there is/was so much inertia with replacing various frame based TDM/SDH/SoNET systems with packet based solutions (this was whole reason why ATM was developed as it is sort of in between of these extremes but still does not have latency and QoS characteristics that were originally sought).
These systems are designed to either induce constant latency (TDM, one frame period per switch) or induce lowest possible latency (SoNET/SDH with very clever mechanism to construct frames without knowing their full content and decoding them incrementally, which gives best case latency of few bit times + latency of any fiber interface, worst case comparable with traditional TDM).
The problem with latency and telephony is two fold: echo cancellation for single pair analog phones (phone itself only substracts outgoing signal from incoming so anything that is echoed from other side of line goes through) and maximal latency for comfortable conversation (which is surprisingly small). First problem can be solved with smarter interfaces that do echo-cancellation in some DSP (some kind of license for doing that is major part of cost of analog interface cards for open source PC based softswitches), second not so much.
So in all you can reasonably jam remotely (on distances of say less than 2000km) using this infrastructure and using it is probably few orders of magnitude cheaper (renting channels in SDH/SoNET optical networks is actually surprisingly cheap) than laying special fibres.
Given that the speed of light is approx 300,000 kilometers per second and that latency of greater than about 20ms starts to interfere with most musicians (and probably less for skilled artists), you're looking at a hard limit of about 6000km for a jam session, and that's assuming zero latency in the rest of the signal chain. In practice anything over 3000km/2000 mi is likely to be a problem, so live jams between NYC and Tokyo are probably never happening unlesss we come up with some clever way to exploit spooky action at a distance.
Laying this new fiber would probably be economically viable for low latency trading. They're already building microwave networks to take advantage of the spread between speed of light in air and glass (http://www.securitiestechnologymonitor.com/news/mckay-brothe...). Based on the fact that most long haul networks use fiber rather than microwave, I would assume that fiber is more economical than microwave.
36 comments
[ 3.2 ms ] story [ 79.2 ms ] threadIf by "all but eliminating" they mean "reducing by 31%".
There's a hard limit to how low latency can go. We've been close to it for years. So a hop takes 100 ms instead of 150 ms? They're still in the same ballpark. To have a revolutionary impact, it would have to drop the latency an order of magnitude, which, as far as we understand our universe, is impossible.
I guess the real application is over short distances. I'd love to hear the ideal usecase that benefits from this improvement.
Still not the case. The state-of-the-art networking technologies are expensive (300K+/yr per data center plus inter-facility connectivity running into the 1-5M/yr range)
Lets say your application needs <50ms latency, low enough that your users can't notice. Today, you'll something around 10 datacenters in the USA/Canada to achieve this.
If cut that down by 31%, you might only need 6. (or less, this would take some real math to figure out).
But the impact is pretty large in the long run, if it can really cut long haul latencies by 31%.
At only 37 channels, compared to the 80 that more traditional fiber can carry using DWDM, it also carries less than half the capacity. And that's assuming that it's capable of the same bandwidth capacity per channel; the article states 40Gbps per channel, whereas you can already go as high as 100Gbps over a single wave using DWDM, which means it may only be able to carry 18.5% as much.
[1] http://www.forbes.com/forbes/2010/0927/outfront-netscape-jim...
http://www.extremetech.com/extreme/122989-1-5-billion-the-co...
Failures can cause exchange crashes big time. This has nearly happened a few times before.
Risk is centralised and multiplied.
HFT contributes to volatility.
Hurts long term investment by human brokers.
- Can't failures be counteracted by having failsafe measures? For example, if the market goes down more than 5% in a few-minute period, the market temporarily shuts down for a cooloff period of several hours to diagnose what's happening.
- Isn't HFT irrelevant to long term investment, given that it erodes away short-term profit, leading to accurate pricing of the stock?
A failsafe/circuit breaker is a possibility but shutting down the market will affect confidence. I'd prefer to see a market operate on a tick basis I.e all trades for a tick are transacted in one go with a window between each tick and assertions before and after the tick processing.
Fair point with your second comment.
Since market participants assume that the volume they see is liquidity, when they try to execute larger orders, and BAM! (this happens with single large orders as well as a dump of 100 lots) the floor just vanishes from them as machines just turn off for some time while someone liquidates a position and tries to find ACTUAL buyers.
http://en.wikipedia.org/wiki/Series_of_tubes
If you think about a recording studio experience, you usually lay down a rough basic track and overdub on top of it. In that case, latency doesn't really matter as much as good tools for collaboration and iteration.
These systems are designed to either induce constant latency (TDM, one frame period per switch) or induce lowest possible latency (SoNET/SDH with very clever mechanism to construct frames without knowing their full content and decoding them incrementally, which gives best case latency of few bit times + latency of any fiber interface, worst case comparable with traditional TDM).
The problem with latency and telephony is two fold: echo cancellation for single pair analog phones (phone itself only substracts outgoing signal from incoming so anything that is echoed from other side of line goes through) and maximal latency for comfortable conversation (which is surprisingly small). First problem can be solved with smarter interfaces that do echo-cancellation in some DSP (some kind of license for doing that is major part of cost of analog interface cards for open source PC based softswitches), second not so much.
So in all you can reasonably jam remotely (on distances of say less than 2000km) using this infrastructure and using it is probably few orders of magnitude cheaper (renting channels in SDH/SoNET optical networks is actually surprisingly cheap) than laying special fibres.