Increasing the degree of parallelism using speculative execution in task-based runtime systems.

Bérenger Bramas

doi:10.7717/peerj-cs.183

Abstract

Task-based programming models have demonstrated their efficiency in the development of scientific applications on modern high-performance platforms. They allow delegation of the management of parallelization to the runtime system (RS), which is in charge of the data coherency, the scheduling, and the assignment of the work to the computational units. However, some applications have a limited degree of parallelism such that no matter how efficient the RS implementation, they may not scale on modern multicore CPUs. In this paper, we propose using speculation to unleash the parallelism when it is uncertain if some tasks will modify data, and we formalize a new methodology to enable speculative execution in a graph of tasks. This description is partially implemented in our new C++ RS called SPETABARU, which is capable of executing tasks in advance if some others are not certain to modify the data. We study the behavior of our approach to compute Monte Carlo and replica exchange Monte Carlo simulations.

Highlights

Parallel CPUs are everywhere, from mobile phones to high-performance computing nodes
In ‘Motivation Example: Monte Carlo Simulations’, we describe the MC and replica exchange Monte Carlo (REMC) simulations and discuss how they are usually parallelized as a motivation to our work
We provided the first results of using speculation in task-based runtime system (RS)’s

Summary

INTRODUCTION

Parallel CPUs are everywhere, from mobile phones to high-performance computing nodes. Some write dependencies are not true because a task might not modify the data, but this cannot be known in advance This motivated us to look to the side of speculative execution, with a primary objective of improving the scalability of the MC/REMC algorithms. It continues by computing the new energy with this domain that moved at line 11. At line 13, we use the Metropolis formula to decide, based on the energy difference and temperature, if the change has to be accepted. // Accept the move (or do nothing) if random_01() ≤ metropolis(new_energy, energy, temperature) domains ← replace(domains, d, new_d) energy ← new_energy end end end Replica exchange Monte Carlo simulation (REMC).

BACKGROUND

Findings

CONCLUSION