A New Theory of Deadlock-Free Adaptive Routing in Wormhole Networks
IEEE Transactions on Parallel and Distributed Systems
Performance of the CRAY T3E multiprocessor
SC '97 Proceedings of the 1997 ACM/IEEE conference on Supercomputing
Interconnection Networks: An Engineering Approach
Interconnection Networks: An Engineering Approach
Tiny Tera: A Packet Switch Core
IEEE Micro
The Reliable Router: A Reliable and High-Performance Communication Substrate for Parallel Computers
PCRCW '94 Proceedings of the First International Workshop on Parallel Computer Routing and Communication
Optimal fault-tolerant embedding of paths in twisted cubes
Journal of Parallel and Distributed Computing
Enhanced fault tolerant routing algorithms using a concept of "balanced ring"
Journal of Systems Architecture: the EUROMICRO Journal
rHALB: a new load-balanced routing algorithm for k-ary n-cube networks
APPT'07 Proceedings of the 7th international conference on Advanced parallel processing technologies
Journal of Computer Systems, Networks, and Communications
Building a terabit router with XD networks
ACSAC'05 Proceedings of the 10th Asia-Pacific conference on Advances in Computer Systems Architecture
Hi-index | 0.00 |
More and more attention is focused on direct interconnection networks when designing the switching fabrics in the terabit routers. Various switching mechanisms are proposed for multi-computer systems, which also rely on direct interconnection networks between processors to support the messages passing mechanism. But it remains unknown which one is more suitable for fabrics in the terabit routers. Based on the requirements of terabit class routers we made analysis and simulations on various switching mechanisms, such as store and forward, wormhole switching, virtual cut through switching and pipelined circuit switching. The results show that virtual cut through exhibits superior performance characteristics over other switching mechanisms under various conditions. Simulations of the performances of virtual cut through shows that larger buffer, longer flit and more virtual channels help to sustain higher throughput at the cost of increasing latency.