- Silicon Valley's 19 Coolest Places to Work
- Is Windows 8 Development Worth the Trouble?
- 8 Books Every IT Leader Should Read This Year
- 10 Hot Hadoop Startups to Watch
Network World - Graduate students at the MIT Computer Science and Artificial Intelligence Laboratory showed off their latest research at the university's Wireless retreat on Monday, outlining software-defined MIMO, machine-generated TCP optimization, and a localized wireless networking technique that works through sound.
Swarun Kumar's presentation on OpenRF – a Wi-Fi architecture designed to allow multiple access points to avoid mutual interference and focus signals on active clients – detailed how commodity hardware can be used to take advantage of features otherwise restricted to more specialized devices.
There were several constraints in the 802.11n wireless standard that had to be overcome, Kumar said, including a limitation on the total number of bits per subcarrier signal that could be manipulated, as well as restricting that manipulation to one out of every two such signals.
Simply disabling the Carrier Sense restrictions, however, proved an incomplete solution.
“Access points often send these beacon packets, which are meant for all clients in a network … you cannot null them at any point if you're a client. Unfortunately, these packets will now collide" in the absence of Carrier Sense, he said.
The solution – which involved two separate transmit queues – enabled OpenRF to automatically apply its optimal settings across multiple access points, distributing the computational workload across the access points, rather than having to rely on a beefy central controller.
Kumar said the system can boost TCP throughput by a factor of 1.6 compared to bare-bones 802.11n.
Keith Winstein attacked the problem of TCP throughput slightly differently, however. Using a specialized algorithm called Remy – into which users can simply input network parameters and desired performance standards – he said that networks can essentially determine the best ways to configure themselves on their own.
“So these are the inputs, and the output is a congestion control algorithm,” he said. “Now this is not an easy process – this is replacing a human protocol designer. Now, it costs like $10 to get a new protocol on Amazon EC2.”
Remy works via the heuristic principle of concentrating its efforts on the use cases where a small change in the rules results in a major change in the outcome, allowing it to optimize effectively and to shift gears quickly if network conditions change.