Genetic algorithm based task reordering to improve the performance of batch scheduled massively parallel scientific applications

Abstract

SUMMARYThe growth in size of networked high performance computers along with novel accelerator-based node architectures has further emphasized the importance of communication efficiency in high performance computing. The world's largest high performance computers are usually operated as shared user facilities due to the costs of acquisition and operation. Applications are scheduled for execution in a shared environment and are placed on nodes that are not necessarily contiguous on the interconnect. Furthermore, the placement of tasks on the nodes allocated by the scheduler is sub-optimal, leading to performance loss and variability. Here, we investigate the impact of task placement on the performance of two massively parallel application codes on the Titan supercomputer, a turbulent combustion flow solver (S3D) and a molecular dynamics code (LAMMPS). Benchmark studies show a significant deviation from ideal weak scaling and variability in performance. The inter-task communication distance was determined to be one of the significant contributors to the performance degradation and variability. A genetic algorithm-based parallel optimization technique was used to optimize the task ordering. This technique provides an improved placement of the tasks on the nodes, taking into account the application's communication topology and the system interconnect topology. Application benchmarks after task reordering through genetic algorithm show a significant improvement in performance and reduction in variability, thereby enabling the applications to achieve better time to solution and scalability on Titan during production.

Parallel accelerated vector similarity calculations for genomics applications

“…However, we believe code modifications are possible which would make it unnecessary to use a random permutation of nodes: by an adjustment of the code it should be possible to recast the communication pattern as a nearest neighbor communication and then use known methods to map the communication pattern optimally to the network, see, e.g. [33]; this will be a topic of future study. In any case, as with other parallel applications, optimizing communications in a multiuser environment is challenging insofar as the network bandwidth is shared by other users and furthermore it is not always possible for a user to reserve a communication-optimal subset of nodes for job execution.…”

Communication Characterization and Optimization of Applications Using Topology-Aware Task Mapping on Large Supercomputers

“…Sankaran et al [33] used a genetic algorithms for optimizing the mapping for two large-scale parallel S3D and LAMMPS on the Cray XK7 machine. Bhanot et al [13] used simulated annealing to optimize task layout of parallel applications SAGE and UMT2000 on the BlueGene/L machine.…”

A Generic Adaptive Multi-Gene-Set Genetic Algorithm (AMGA)

“…The search space is reduced by generating a full-active schedule that satisfies precedence constraints, a neighborhood search is applied to exploit the search space for better solutions and to enhance the GA. Simulation suggests sustainability of this hybrid GA in solving JSSP. Sankaran et al [21] proposed a GA based parallel optimization technique aiming to improve the performance of batch schedule of two massively parallel application codes; a turbulent combustion flow solver (S3D) and a molecular dynamics code (LAMMPS). Experiments have shown a significant deviation from ideal weak scaling and variability in performance.…”