Enhancing the performance of malleable MPI applications by using performance-aware dynamic reconfiguration

Abstract

The work in this paper focuses on providing malleability to MPI applications by using a novel performance-aware dynamic reconfiguration technique. This paper describes the design and implementation of Flex-MPI, an MPI library extension which can automatically monitor and predict the performance of applications, balance and redistribute the workload, and reconfigure the application at runtime by changing the number of processes. Unlike existent approaches, our reconfiguring policy is guided by user-defined performance criteria. We focus on iterative SPMD programs, a class of applications with critical mass within the scientific community. Extensive experiments show that Flex-MPI can improve the performance, parallel efficiency, and cost-efficiency of MPI programs with a minimal effort from the programmer.

Efficient Scalable Computing through Flexible Applications and Adaptive Workloads

“…Another similar approach is found in [14], where a performance-aware framework based on the Flex-MPI library [15] is presented. That work leverages job reconfiguration in order to expand/shrink a job targeting execution performance.…”

Improving the Energy Efficiency of MPI Applications by Means of Malleability

“…Flex-MPI consists of several modules. For a more detailed description, please refer to [10]. The paper contributions are highlighted in red, which are either newly created or extended with new functionalities.…”

“…Algorithm 2 is an extension of the algorithm depicted in [10]. The algorithm consists of three phases.…”

“…To implement our solution, we have extended Flex-MPI [10] with new functionalities. Flex-MPI provides dynamic malleability to iterative SPMD MPI applications and performs runtime monitoring via hardware performance counters.…”

Malleable parallelism with minimal effort for maximal throughput and maximal hardware load.

“…Others employ more intricate methods, e.g., the ReShape framework [10,11], also built on-top of MPI-2, or the flexMPI framework [12,13] built on MPI-3, or the proposed ULFM standard API [8] that perhaps will be included in the future MPI-4. Yet other studies [14][15][16][17] employed various application management systems (Charmm++, EasyGrid, Nanos++) as additional software layer between the application and the scheduler.…”