The method, called “3D beamforming,” was created by scientists at the University of California, Santa Barbara. The basic premise is that data center traffic gets tied up at key points during the day because of congestion associated with short bursts of activity. In other words, the key speed constraint is not the total amount of network bandwidth, but rather how much is consumed in those short bursts.
Typically, data centre managers try to get around bandwidth issues by maintaining the maximum bandwidth needed to handle peak periods. The rest of the time, though, bandwidth can sit idle. So, rather than continuing to add more infrastructure that’s used only occasionally, the idea is to send those bursts of traffic directly between the servers that need the information — avoiding competition with other servers and applications — and to do so at top speeds.
Scientists used mostly off-the-shelf gear to set up the system that creates and transmits the 60GHz beams, according to Heather (Haitao) Zhang, one of six primary researchers. The design they created bounces data over server racks, off metal plates on the ceiling, to specially designed antennas on each server.
Another advantage: The wireless links can be turned off and on as needed, saving bandwidth in the process. But to make sure those connections can be established, broken and remade at will, “the next step is to develop network protocols and management techniques,” Zhang said. “We are currently working on these,” she noted, adding that she expects this phase of the project to take a couple of years.