Machine learning based online performance prediction for runtime parallelization and task scheduling

Abstract

Abstract-With the emerging many-core paradigm, parallel programming must extend beyond its traditional realm of scientific applications. Converting existing sequential applications as well as developing next-generation software requires assistance from hardware, compilers and runtime systems to exploit parallelism transparently within applications. These systems must decompose applications into tasks that can be executed in parallel and then schedule those tasks to minimize load imbalance. However, many systems lack a priori knowledge about the execution time of all tasks to perform effective load balancing with low scheduling overhead.In this paper, we approach this fundamental problem using machine learning techniques first to generate performance models for all tasks and then applying those models to perform automatic performance prediction across program executions. We also extend an existing scheduling algorithm to use generated task cost estimates for online task partitioning and scheduling. We implement the above techniques in the pR framework, which transparently parallelizes scripts in the popular R language, and evaluate their performance and overhead with both a realworld application and a large number of synthetic representative test scripts. Our experimental results show that our proposed approach significantly improves task partitioning and scheduling, with maximum improvements of 21.8%, 40.3% and 22.1% and average improvements of 15.9%, 16.9% and 4.2% for LMM (a real R application) and synthetic test cases with independent and dependent tasks, respectively.

Scalable black-box prediction models for multi-dimensional adaptation on NUMA multi-cores

“…Li, et al [26], similarly to our work, use ANNs as models for online performance prediction, that they apply to task partitioning and scheduling for HPC clusters. Their work does not consider NUMA architectures, core and data layout effects, or power metrics.…”

“…Li, et al, [26] and Ipek, et al, [27] use artificial neural networks (ANNs) as blackbox models for microarchitectural design space exploration. Li, et al [26], similarly to our work, use ANNs as models for online performance prediction, that they apply to task partitioning and scheduling for HPC clusters.…”

Scheduling Optimization of Parallel Linear Algebra Algorithms Using Supervised Learning

“…They measured assignments, branches, and loops at run-time using dynamic analysis of the program. In [22], non-deterministic features were measured. The variables ar, ot were represented using performance counters: number of CPU cycles, number of cache misses, cache accesses for the last cache level, and number of level one cache hits.…”

Machine learning‐based auto‐tuning for enhanced performance portability of OpenCL applications

“…In supervised learning, on the other hand, the algorithm is provided Best tuning params. 13,15,20,21,28,29 7,29 24,30,31 Work distribution 25,34 33,34 Power 23,27 Output size 17 between the input and output variables of interest. These data are then given to a machine learning algorithm that builds a model.…”