What do we do when our network slows to a crawl? Throw more money at the problem.
But a potential breakthrough from the Massachusetts Institute of Technology may change the way we approach networking in data centres. MIT researchers are working on a network management system, called Fastpass, which could significantly reduce data centre lag times.
There’s a reason why the researchers gave it that name: In experiments, Fastpass reduced the average queue length of routers in a Facebook data centre by as much as 99.6 per cent. And latency dropped from 3.56 microseconds to 0.23 microseconds.
That could pretty much do away with queues, which plague today’s congested networks.
Much like traffic at a Tim Hortons’ drive-through window at morning rush hour, when packets of data show up at a router at the same time, they end up in a queue. And the longer that queue gets, the longer it takes for those packets to get where they need to go.
Since traffic is so unpredictable, you’ve probably had to invest in fatter pipes to handle heavy loads and put a queue in each switch.
With Fastpass, switches don’t decide where those packets will go; instead, a central server, or arbiter, is now in charge. It checks out the entire data centre environment and then schedules packets in a more efficient way.
This “zero-queue” data network architecture determines the time a packet should be transmitted and what path to use.
In the Facebook experiment, an arbiter with eight cores 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 centre with gigabit-speed links running at full speed, according to the co-authors of an MIT paper on Fastpass.
Sure, this would probably make videos on YouTube load faster, but that’s not the motivation behind the research. Fastpass, according to the paper, is about simplifying the network and shrinking the amount of bandwidth required in data centres. Ultimately, it could change the way we write web and mobile apps.
Hari Balakrishnan, a professor and one of the co-authors of the paper, says Fastpass could be implemented in dedicated server clusters or built into specialized chips. And it has the potential to “reduce the administrative cost and equipment costs and pain and suffering to provide good service to the users.”
Of course, it will take awhile for widespread adoption. Fastpass is available as open source software, but it’s not production-ready code; researchers expect it will show up in production data centres in about two years.
Still, we could be at the start of a new era — and the dream of a zero-queue data centre could be reality much sooner than we think.
Image thanks to Nietnagel