Network Design in the Big-Data Age

Today, the design of next generation networks is dictated by the characteristics of big-data applications that generate large volumes of data traffic.  The sheer size and dynamic nature of these workloads drive network design tradeoffs.  For example, cellular network providers are scrambling for additional spectrum to support heavy network usage on mobile devices, and batched data processing jobs are straining network capacity in data center networks.  In both cases, network architects must find novel ways to support high traffic demands that vary significantly with time, often without predictable patterns.

In this talk, I will present key results from our efforts to tackle the issue of traffic dynamics in both next generation wireless networks and data centers.  First, I will describe my work on addressing dynamic traffic demands in wireless networks using dynamic auctions for radio spectrum.  Highly dynamic spectrum usage calls for on-demand auctions for fine-grain spectrum leases.  I will show that existing auction designs fail due to complex interference constraints in the wireless environment.  Our solution is an eBay-like spectrum marketplace that auctions spectrum efficiently while achieving economic-robustness.  Second, I will describe challenges in dealing with dynamic traffic hotspots in data center networks, where our solution is to use 3-D flexible wireless interconnects to augment wired networks.  I will identify key challenges of using wireless in data centers: link blockage and radio interference, and then present 3D beamforming as a new wireless primitive to address them.  Finally, I will discuss my plans for both short-term and long-term future work.