Determining the MPP LS-DYNA Communication and Computation Costs with the 3-Vehicle Collision Model and the Infiniband Interconnect
The least square error approach applied to LS-DYNA communication and computation costs for the Neon model was shown previously by this author to be useful in predicting performance on a given interconnect of known ping- pong latency and bandwidth, and both Gigabit Ethernet and HyperFabric 2 results were presented. In this paper, this prediction method is applied to a much larger public-domain crash model, the 3-vehicle collision model, to determine communication and computation costs for models representative of the most demanding requirements. Furthermore, the result is verified against the new, high-speed low- latency Infiniband interconnect. Users of this method may perform trade-off analysis for optimum hardware configuration decisions without the need for extensive benchmark testing.
https://www.dynalook.com/conferences/international-conf-2004/12-3.pdf/view
https://www.dynalook.com/@@site-logo/DYNAlook-Logo480x80.png
Determining the MPP LS-DYNA Communication and Computation Costs with the 3-Vehicle Collision Model and the Infiniband Interconnect
The least square error approach applied to LS-DYNA communication and computation costs for the Neon model was shown previously by this author to be useful in predicting performance on a given interconnect of known ping- pong latency and bandwidth, and both Gigabit Ethernet and HyperFabric 2 results were presented. In this paper, this prediction method is applied to a much larger public-domain crash model, the 3-vehicle collision model, to determine communication and computation costs for models representative of the most demanding requirements. Furthermore, the result is verified against the new, high-speed low- latency Infiniband interconnect. Users of this method may perform trade-off analysis for optimum hardware configuration decisions without the need for extensive benchmark testing.
12-3.pdf
— 60.0 KB
