Researchers at MIT are developing a “Fastpass” network management system that promises to significantly reduce the latency between servers in hyperscale data centers.
According to a university blog, the MIT Fastpass system uses a central server called an “arbiter” to decide which nodes in the data center network may send data to other nodes in a designated time slot. An arbiter based on the latest multicore silicon reportedly can keep up with a network carrying over 2 terabits of traffic. The article estimates the system could reduce
latency in a Facebook data center by 99%.
The MIT researchers plans to present their work at an upcoming conference in August.
