Workshop: Sub-Microsecond Optical Circuit Switched Networks for Data Centers and HPCs
Abstract: There have been over two decades of research on optical switched technologies for Data Centers and HPCs. However there is still no commercial uptake and adoption by Cloud and HPC providers. This presentation will delve into why is that and elaborate on the roadblocks, opportunities and choices to address the Moore’s Law slow down and replace electronic switching. It will explain the reasons why optical circuit switching rather than optical packet switching is the best way forward for Data Centers and HPCs and what are the main challenges. PULSE, a scalable network architecture that can reconfigure optical time-slotted circuits in nanoseconds will be introduced to support heterogeneous high performance systems . An insight on technologies we are developing for sub-microsecond network scheduling and sub-nanosecond switching that aim to maximise throughout and minimise latency across thousands of end points (Servers, CPUs, accelerators, memory, etc.) will be provided. Crucially, I’ll present how the network can offer zero data loss, without in-network a) buffering, b) active switching and routing, and c) network header addressing and processing to minimize complexity, and crucially to consume very low power. Furthermore, the system also aims to inherently support 1-to-1, 1-to-N, N-to-N and N-to-1 connectivity in a synchronous manner without the need for data replication for multi/broad -casting, currently not possible. This is key to support diverse workloads such as storage caching, large-scale database lookups, training distributed DNNs, parallel computing that use communication primitives such as allreduce, broadcast and reduce, gather and scatter, all-to-all among others.