Enabling Quasi-Static Reconfigurable Networks With Robust Topology Engineering

Many optical circuit switched data center networks (DCN) have been proposed in the last decade to attain higher capacity and topology reconfigurability, though commercial adoption of these architectures minimal. One major challenge face is difficulty handling uncertain traffic demands using switches (OCS) with high switching latency. Prior works generally focused on developing fast-switching OCS prototypes quickly react variations through frequent reconfigurations. This approach, however, adds tremendous complexity overhead control plane, raises barrier for networks. We propose, a robust routing optimization framework reconfigurable centers. co-optimizes based convex set matrices, offers strict throughput guarantees any future matrices bounded by set. For bursty that are unbounded set, we employ desensitization technique reduce performance hit. enables generate solutions capable unexpected changes without relying Our extensive evaluations Facebook’s production DCN traces show that, even daily reconfigurations which could be realized current MEMS-based OCSs from Calient Technologies, achieves about 20% lower max link utilization, 32% average hop count compared cost-equivalent static topologies. work shows topologies DCNs feasible fast OCSs.