A BIT of congestion can quickly lead to a big slowdown. This is as true of the internet as it is of driving. When a file is transmitted over a network it is divided into packets, or chunks of data, which are re-asse
mbled in the same order at the other end. Many networking protocols recognise when a packet is missing, as they are numbered to keep track. Responses vary, but protocols which guarantee a packet’s delivery, like the ubiquitous TCP used on local networks and across the internet, rely on acknowledgement of each packet’s receipt. If a receiver fails to do so, or takes too long to respond, the sender will dispatch missing bundles again. This can have huge effects on throughput—its net speed after addressing information, metadata and other factors are removed.
Laila Partridge, the boss of Code On, offers a solution. Her firm is in the process of licensing software and intellectual property to networking companies. Corporate partners will be unveiled in a few months but some academics whose work contributed to the technology let the kimono slip recently during a shindig at the Massachusetts Institute of Techology. They showed how a change in transmitting packets could produce remarkably high-quality streaming video over poor network links.
Where the classic approach slices long runs of data into packets which comprise just a piece of the whole, Code On converts each packet into an algebraic equation with a randomly generated coefficient. Clever maths allows the original packet to be recreated from the formula’s constituent parts. Cleverer still, Code On’s approach recombines elements of formulae from multiple packets into each chunk sent, allowing the addition of redundancy. This increases the raw amount of data sent. But it improves the odds that a missing packet may be reconstructed from neighbouring ones that were received. Moreover, redundancy can be increased or decreased depending on the network’s quality. The improved throughput vastly exceeds the additional data load.
Crucially, Code On’s solution does not require rerigging the internet. It simply fools TCP software in connected devices into thinking they are exchanging traditional packets. The company’s software mimics TCP’s responses, preventing TCP from triggering its slowdowns without breaking compatibility with the rest of the internet. Only when the redundancy is insufficient or network quality too poor would Code On’s protocol give up and let TCP take over.
Ms Partridge says that while both sides of a coded communication need to have its software enabled in the same fashion, its approach works in any kind of network protocol in which dropped packets hobble performance. (Protocols that do not promise every packet will arrive are often used for voice calls and video streaming, where the stream is more important than any single bit in it.) As a consequence, significant improvements could be had by changes made to desktop, mobile and router operating systems which typically handle TCP tasks, rather than to immutable chips.
Code On’s techniques will probably first appear in applications like video-streaming software, in which a company such as Netflix controls both its own servers and the client software. Various techniques are already in place to buffer video, handle missing pieces and reconstruct from errors. But Code On’s more general approach could do more with less effort. Other early users are likely to include internet service providers and their hardware suppliers who could upgrade both central-office systems and the software in customers’ broadband modems. The change would be invisible to consumers, other than in improved performance, especially during hours of peak use.
The approach also allows for the easier mixing of multiple networking streams at once. Smartphones, for instance, switch seamlessly between Wi-Fi and mobile broadband. But the packets flow from one source or another, never from both at once. Ms Partridge says Code On’s protocols allows both (or even many) radios to be used at the same time, which will be relevant in upcoming mobiles with more advanced radio technology, where the balance between Wi-Fi and mobile data would vary depending on network quality, cost and other factors. The device, and its user, would be none the wiser.
Powered by Facebook Comments