MIT researchers think they have a way to speed packet forwarding in networks, which could cut the amount of bandwidth IT administrators have to buy in their data centres.
The bad news is it likely won’t be commercialized for two years. However, the potential is enormous.
The work of the researchers, which incudes a staffer from Facebook is to be presented at Association for Computing Machinery conference next month.
Computerworld U.S. says the computer scientists have created software called Fastpass which uses parallel computing to help switches decide when each packet of data can be forwarded to a destination. Today packets spend a lot of time in memory-intensive lineups waiting for the switch to make a decision.
Fastpass gives that job to a central server which can schedule packets in a more efficient way, the report says.
In a way, the idea is an extension of software-defined networking (SDN).
“By making all packets arrive on time, Fastpass can also save network architects from having to overprovision data center links for unpredictable bursts of traffic,” says the report. “As the number of users and the volume of data grows, it should be easier to keep up. There’s a similar benefit for developers of distributed applications, which split problems up and send them to different servers around a network for answers.
How big a difference can it make? In experiments in a Facebook data center, Computerworld says, Fastpass cut the average length of a queue by 99.6 percent, the researchers say. Latency, or the delay between requesting and receiving an item, went from 3.56 microseconds to 0.23 microseconds.
That may not seem a lot, but when millions of packets are on the line it adds up.
In the test, an eight core Fastpass server was able to make decisions for a network carrying 2.2 terabits of data per second, which is equal to a 2,000-server data center with gigabit-speed links running at full speed, MIT said.
The question is whether the solution can be practically applied in most data centres or is it only applicable in service-provider networks.