Processor assignment in heterogeneous parallel architectures

Abstract

It has been already demozstrated that cost-effective multiprocessor designs may be obtained by combining in the same architecture processors of diflerent speeds (heterogeneous architecture) so that the serial and critical portions of the application may benefit from a fast single processor. This paper presents a systematic way to build static heuristic scheduling algorithms for such environments. Several algorithms are proposed and their performance are compared through simulation. One of the proposed algorithms is shown to achieve substantial performance gains as the degree of heterogeneity of the architecture increases.

Multiprocessor scheduling under precedence constraints: Polyhedral results

“…When |S| = |T | the above proof shows that inequality (24) can be obtained by nonnegative linear combinations of other valid inequalities, and therefore it does not define a facet of P PLO (D n ). Now assume that s = |S| < |T |.…”

“…The results attained so far, show that for any feasible point (z, y) in F, the contribution of y to the left-hand side of z + y = 0 is always given by u∈N u . This means that with respect to inequality (24), the machine to which the vertices are assigned is irrelevant. Therefore, the relationships among the coefficients of the vectors , and 0 that still have to be established to prove that inequality (24) defines a facet can be obtained using exactly the same points that are given in the proof of Theorem 11.…”

“…In a heterogeneous multiprocessor system, we not only have to determine how many, but also which processors should be allocated to an application and which tasks will be assigned to each processor. Greedy algorithms for processor assignment of parallel applications modeled by task precedence graphs in heterogeneous multiprocessor architectures were proposed by Menascé and Porto [24], while Porto and Ribeiro [27,28] studied sequential and parallel algorithms based on the tabu search metaheuristic. Porto et al [26] presented a detailed analysis of the solutions obtained by this parallel tabu search algorithm, using a broad set of test instances corresponding to real-size and realistic problems and showing that it leads to much better solutions than the greedy algorithm.…”

A new formulation for scheduling unrelated processor under precedence constraints

“…In a heterogeneous multiprocessor system, we not only have to determine how many, but also which processors should be allocated to an application, as well as which processors are going to be assigned to each task. Greedy algorithms for processor assignment of parallel applications modeled by task precedence graphs in heterogeneous multiprocessor architectures have been proposed by Menascé and Porto [9], while Porto and Ribeiro [11,12] have studied sequential and parallel algorithms based on the tabu search metaheuristic.…”

Parallelizing the Microcanonical Optimization Metaheuristic: A Case Study for the Task Scheduling Problem

“…They have becn implemented on a network of workstations using lhe MPI communication library, and an evaluation of lhe quality of lhe solulions generaled has been perfonned for different sets of lhe algorithm paramelers. The solulion qualily has been measured by the makespan reduction achieved in comparison with the best greedy algorithm, and with tabu search, for lhe same problem instances [7,10]. The condítions under which lhe new algorithm is able to show a superior performance are then highlighted by our preliminary results.…”