SSA NoC Performance Evaluation
Task 2.2 Validation, performance and scalability analysis of SSAs in Intel’s SCC using large biomolecular networks
The performance analysis of the parallel stochastic simulation framework we have developed was done by simulating biomodels of varying complexity in SSIP and MSIP mode of parallel execution on the Intel SCC NoC processor. In addition, in order to compare performance scalability of many-core vs. multi-core CPUs on the problem we also run the same experiments on a high-end multi-core CPU, namely the quad processor Intel Core i7 last generation. The results and analysis of performance scaling suggest that the future is definitely many-core! Although the 48 cores available to the Intel SCC NoC CPU are limited in terms of capacities and power (Pentium 4, P54C, cores) compared with the 4 cores available to the Intel Core i7 CPU, their larger number and the very fast interconnection network on chip allowed a greater speedup and hence efficiency while performing stochastic simulations of large size biomodels.
During the computational experiments we evaluated the throughput and performance of the parallel computing framework when used to run the same biomodel on an increasing number of cores. Specifically for each experiment we measured the simulation execution time, the overall number of Reaction Cycles (RC) of the biomodel performed during all iterations of the simulation, and calculated using them the throughput (in millions of reaction cycles per second - MRC/sec), the overall performance (in millions reactions per second - MR/sec), the speedup (S) factor that is obtained relatively to a single core, and the efficiency (E), which is the ratio of the speedup to the number of the cores used in the simulation, for a variety of different configurations.
All stochastic simulation computational experiments were performed both in the SSIP and in the MSIP modes of parallel operation using the developed software framework appropriately configured. The first basis for benchmarking (baseline) was determined by running the same simulation in a single core of the Intel SCC NoC processor, i.e. an Intel Pentium (P54C) core operating at 800MHz under the SCC Linux operating system. The second baseline was determined by running the same model by using the software framework that we developed, but now executed in a very powerful PC workstation having an Intel Core i7 4790K CPU, running at 4GHz (with all cores active - not Turbo), 32 GB RAM and operating system GNU / Linux.
In carrying out stochastic simulation experiments we observed that as the number of reactions of the model increases, both the performance and efficiency are scaling up accordingly, approaching almost ideal values (linear speedup) when possessing highly complex biomodels, i.e. networks with thousands of molecular species and biochemical reactions.
D2.2 Performance scalability evaluation when simulating biomodels of increasing complexity on the SCC
The detailed results of the research can be found in the publication:
Elias Kouskoumvekakis, Dimitrios Soudris, Elias S. Manolakos: Many-core CPUs can deliver scalable performance to stochastic simulations of large-scale biochemical reaction networks. In Proceedings of the International Conference on High Performance Computing & Simulation, HPCS 2015, Amsterdam, Netherlands, July 20-24, 2015, IEEE 2015, ISBN 978-1-4673-7812-3 HPCS 2015:517-524
presented by the first author (MSc student) as an oral presentation in the leading International HPCS conference. It is noteworthy that the above publication was a candidate for the best / outstanding paper award. Only 9 out of a total of more than 100 publications presented at the conference reached this distinction, reaffirming the important and interesting work carried out overall by our team in this project.