System simulation methodology of optical interconnects for high-performance computing systems

Abstract

The relentless quest for processing speed in the range of teraflops and beyond has accelerated the need for scalable, parallel, high-performance computing (HPC) systems. To meet this high bandwidth and low power requirements, optical interconnect-based system architectures are being implemented by the HPC industry. While computer-aided design tools have significantly assisted electronic system simulation, the field of system level optoelectronics modeling has lagged behind owing to lack of simulation methodologies and tools. This paper explores the design space of developing OPTISIM, a system level modeling and simulation methodology of optical interconnects for HPC systems. OPTISIM can provide computer architects, designers, and researchers a highly optimized, efficient, and accurate discrete-event environment to test various research hypotheses on HPC systems with power-performance implications. For any given optical interconnect architecture with optical transceivers, wavelength assignment, and traffic patterns, OPTISIM provides end users with network throughput, average latency, power loss, power consumption, and signal strength at the output. The proposed OPTISIM simulation methodology is explained with a case study on the performance of an optical HPC architecture called RAPID.