Blog Post

Latency is a pain, and these researchers say they have the cure

There are two main things that can cause a real-time web service to be not-so-real-time: insufficient bandwidth and excessive latency. Bandwidth is an ISP issue, but latency — the time it takes for your computer to contact a server and/or get a response — is largely down to the way the internet is engineered.

So a team of researchers from around Europe, with some help from the European Commission, Alcatel-Lucent and BT, is aiming to revise a standard data-transport protocol to effectively re-engineer how the internet runs — without the need for pricey new equipment.

It’s a tall order, but they’re deadly serious. And from the vendor side, it’s not hard to see why: Alcatel-Lucent wants to slash latency for better videoconferencing, and BT wants to do the same in order to improve its Radianz Cloud platform, which serves the very time-sensitive financial services industry.

The project is called Reducing Internet Transport Latency (RITE). It quietly kicked off last November but the University of Aberdeen, where some of the researchers are based, has only just started making noise about it — largely because the researchers are about to set off for an Internet Engineering Task Force (IETF) meeting in Florida next month to show off what they’re up to.

In a statement, Professor Gorry Fairhurst said:

“It’s a problem we all notice when you’re using a program like Skype. If anyone else in the house is watching a video at the same time, your video connection becomes jerky and often crashes. This affects gamers who want to play online in real time and companies doing stock training – both end up buying special and expensive internet connections to make these work, but often it’s not more bandwidth that’s needed to go faster – it’s less delay.

We think we can reduce this delay by making a set of small but important changes to the way computers and the network process the internet data.”

So, what sorts of changes are we talking about? Fairhurst told me on Thursday that there are two main strands to the RITE project’s work: revising the core Transmission Control Protocol (TCP, a.k.a the flipside to the Internet Protocol) and changing how network routers handle buffering.

“We’re trying to change TCP so that it works better with thin applications — applications that don’t send a huge amount of data and aren’t really interactive, like media streaming and conference calls,” he said. “We would make a small change to the timer mechanism so that you can recover data when you send a burst and lose part of it — it can take quite a lot of time to recover one lost packet — and we also have to do something to the way the congestion window works.

On the other side, to make this work effectively you have to change the way the routers behave as well. People know routers have lots of memory in them, but TCP tends to fill up all the buffers inside. We’re going to make recommendations on how to avoid buffer blocks — this is more directed at operators than people building PC software, but these things have to be done at the same time.”

Fairhurst noted that the TCP revisions would be piloted in Linux (the team is also working on mechanisms for UDP-based streaming applications). After those Linux patches have been rolled out over the next year, the RITE researchers will try to get the standards community to do its thing and then “hopefully convince Microsoft(s msft), Apple(s aapl) and everyone else” to incorporate the changes into their applications. Google, which has developed the SPDY (pronounced “speedy”) protocol for reducing load time, has already shown interest, he claimed.

Apart from better videoconferencing and faster trading platforms, Fairhurst also suggested that success could “make the internet available to a whole new raft of applications,” such as proper virtual reality.

“It’s a small fix that really came from the gaming community originally,” he added. “Gamers really hate delay.”

Apart from the University of Aberdeen (Scotland), other research facilities involved in RITE include Simula Research Labs (Norway), the University of Oslo (Norway), Karlstad University (Sweden) and the Institut Mines-Telecom (France). The consortium has received just over €3.5 million ($4.7 million) from the European Commission.

6 Responses to “Latency is a pain, and these researchers say they have the cure”

  1. Peter Thompson

    “latency … is largely down to the way the internet is engineered.” – the trouble is, the internet *isn’t* engineered, at least as far as performance is concerned. This is one reason why the internet as it currently stands should be considered just a prototype!

  2. Interesting article. I have read about approach to achieve “better videoconferencing and faster trading platforms” already… Video Conferencing often uses UDP not TPC, and serious trading platforms, like High Frequency Trading, use different methods to resolve latency, they build the trading platform IN the exchange. Hopefully this just helps even more.

  3. They are tackling the right problem, but tinkering with TCP won’t deliver the right answer. You still have non-deterministic failure modes, problems as you saturate the network, a mis-allocation of resources among the flows, and non-stationarity (variability of loss and delay). The underlying issues with TCP/IP aren’t being addressed: work conserving queues & excessing data in transit, trying to manipulate a system of two degrees of freedom (load/loss/delay) using just one (load/loss), non-compositionality of the performance of the queues, lack of adequate mechanisms to trade between the flows. To solve the problem, you need to change the origin of your world view. See for more information. The future is polyservice networks that recognise that demand is heterogeneous, and creates a supply that matches it.