A Comparative Analysis of Deadlock Recovery and Avoidance-Based Routing Algorithms in Wormhole-Switched k-Ary n-Cubes Paper review Reviewer : Nthu CS03 kwc
Outline Introduction Introduction Motivation Motivation Simulation model Simulation model Performance evaluation Performance evaluation Comparison Comparison Previous work Previous work Simulation Simulation Conclusion Conclusion My comment My comment
Introduction Motivation : Motivation : Before routing schemes can be successfully incorporated in future practical multicomputers, it is necessary to develop a clear understanding of the factors, which affects their potential performance.
Introduction ( cont. ) Comparison model : Comparison model : Message destinations are uniformly distributed across the network nodes. Nodes generate traffic independently of each other, and which follows a Poisson process with a mean rate of λ g messages/cycle. Message length is fixed and equal to M flits, each of which requires one-cycle transmission time across a physical channels. The local queue in the source node has infinite capacity. Moreover, messages are transferred to the local PE through the ejection channel as soon as they arrive at their destinations. V virtual channels are used per physical channel. K-ary n-cubes. K-ary n-cubes.
Introduction ( cont. ) Performance evaluation Performance evaluation Message latency Message latency The time required by a message to cross from source to destination. Traffic saturation rate Traffic saturation rate The traffic generation rate when network enters the saturation region. The traffic generation rate when network enters the saturation region.
Comparison Previous work Previous work “ Performance comparison of deadlock recovery and deadlock avoidance routing algorithms in wormhole- switched networks ” in IEE Proceedings-Computers and Digital Techniques, Vol. 150, No. 2, pp , “ Performance comparison of deadlock recovery and deadlock avoidance routing algorithms in wormhole- switched networks ” in IEE Proceedings-Computers and Digital Techniques, Vol. 150, No. 2, pp , 2003.
Simulation Comparison Simulation Algorithm used in these works : Simulation Algorithm used in these works : Previous workCurrent work Deadlock Recovery : Compressionless Routing Deadlock Recovery : Compressionless Routing Software-Based Routing Disha Routing Deadlock Avoidance : Duato ’ s Adaptive Routing Deadlock Avoidance : Duato ’ s Adaptive Routing Test-Bed : Deterministic Routing Test-Bed : none
Simulation Comparison Current work Previous work
Simulation Comparison (cont.) Current work Previous work
Simulation Comparison Current work Previous work
Simulation Comparison Current work Previous work
Conclusion The digest of conclusion is list as follows : The digest of conclusion is list as follows : Networks with a small diameter compressionless routing outperforms both deterministic and Duato ’ s adaptive routing, especially when a few number of virtual channels per physical channels are used. Networks with a small diameter compressionless routing outperforms both deterministic and Duato ’ s adaptive routing, especially when a few number of virtual channels per physical channels are used. Either he network size or the number of virtual channels increases the performance advantages of compressionless routing diminish. Either he network size or the number of virtual channels increases the performance advantages of compressionless routing diminish. Throughput of wormhole-switched k-ary n-cubes is highly sensitive to the congestion channel in the source node. Throughput of wormhole-switched k-ary n-cubes is highly sensitive to the congestion channel in the source node. The high degree of adaptivity offered by a particular instance of k- ary n-cubes, notably the hypercube, enables Duato ’ s adaptive routing to better performance characteristics than compressionless routing when a moderate number of virtual channels are used. The high degree of adaptivity offered by a particular instance of k- ary n-cubes, notably the hypercube, enables Duato ’ s adaptive routing to better performance characteristics than compressionless routing when a moderate number of virtual channels are used.
My Comment Conclusions in current work are overlapped or can be conducted by previous work. ( Section of conclusion in both papers are almost the same. ) Conclusions in current work are overlapped or can be conducted by previous work. ( Section of conclusion in both papers are almost the same. ) Current work is a refined report of previous work. But the topic overlapping and lacking originality makes me feel that it should be done in previous work, not in another analysis report. Current work is a refined report of previous work. But the topic overlapping and lacking originality makes me feel that it should be done in previous work, not in another analysis report.