Optical Switching for High‐Performance Computing

An enormous amount of data is being processed and communicated between nodes in Large‐scale High‐Performance Computing (HPC) systems such as supercomputers and large‐scale data centers. The use of parallelism in HPC increases the exchange of data between nodes, and the performance gain depends on the underlying interconnection architecture. Communication between processes (inter‐process) plays a vital role in HPC in terms of cost and performance. The performance in terms of time and space and the energy consumption of HPC workloads on a cluster‐based architecture are also investigated in this chapter. Improved versions of electronic switches such as multi‐tier high radix have reached the bandwidth requirements at a higher cost and power consumption. Optical switching technology has been employed to overcome this issue. Optical interconnect provides low power consumption with high capacity. Focusing on the optical switching in HPC, this chapter attends to areas from basic optical switching techniques to high‐performance optical switching architectures. The chapter covers various approaches including path provisioning and elasticity, security issues in optical switching, and the investigation and characterization of different HPC optical switching architectures. Finally, this chapter highlights some open challenges of optical switching in HPC, such as minimization of collisions in optical packet switching, low‐latency switching capability, and effective channel modeling for high throughput using queueing principles.