The increasing demand for bandwidth coupled with saturating electrical systems is leading the drive for optics as an interconnect technology. High-performance computing systems (HPCS) consist of a large number of components such as processors, memories, and interconnection links. As the number of components in HPCS increases, the probability of a failure also increases. Therefore, it becomes imperative that the system be fault tolerant to ensure high availability even in the presence of faults. We propose a multidimensional optoelectronic architecture, nD-RAPID (reconfigurable and scalable allphotonic interconnect for distributed and parallel systems), where n can be 1, 2, or 3. nD-RAPID provides high bandwidth, low latency, dynamic reconfiguration, and fault tolerance. While designing the fault-tolerant routing algorithm, we have tried to ensure that it provides optimum performance in the absence of faults, shows minimal degradation in the presence of faults, and can tolerate a reasonable number of faults. In the presence of faults the onboard switching mechanism dynamically reconfigures itself to reroute packets along nonfaulty links. Extensive simulation results are presented that compare nD-RAPID with other popular HPCS topologiesr.
ASJC Scopus subject areas
- Computer Networks and Communications
- Atomic and Molecular Physics, and Optics