Scheduling parallel batch jobs in grids with evolutionary metaheuristics

Abstract

In this paper we propose an efficient offline job scheduling algorithm working in a grid environment that is based on a relatively new evolutionary metaheuristic called generalized extremal optimization (GEO). We compare our experimental results with those obtained using a very popular evolutionary metaheuristic, the genetic algorithm (GA). The scheduling algorithm implies two-stage scheduling. In the first stage, the algorithm allocates jobs to suitable machines of a grid; GEO/GA is used for this purpose. In the second stage, jobs are independently scheduled on each machine; this is performed with a variant of a list scheduling algorithm. Both GEO and GA belong to the class of evolutionary algorithms, but GEO is much simpler and requires the tuning of only one parameter, whereas GA requires the tuning of several parameters. The results of the experimental study show that GEO, despite its simplicity, outperforms the GA in a whole range of scheduling instances and is much easier to use.

Hybrid Metaheuristic Based Offline Parallel Job Scheduling in Heterogeneous Computing Systems

“…Such heuristics produced good results for small problem and system sizes. However, to effectively deal with the NP-hard JSP problem in complex HCS platforms, several meta-heuristic based offline solutions are developed over the years [14][15][16][17] [30]. The major focus of the research was on scheduling independent jobs i.e.…”

“…If a job is non-coallocated one, then PEs from first slowest cluster in the sorted available cluster list will be allocated to the job otherwise PEs across the multiple clusters starting from slowest cluster can be allocated to the job (Line [15][16][17][18][19]. In nutshell, Resource allocation heuristic tends to avoid unnecessary co-allocations to reduce communication slowdown to reduced actual execution time, by allocating single cluster PEs and tries to allocate low computation power resources to jobs to further reduce energy consumption.…”

“…This updated velocity will help the individuals in PSO phase to fly towards global best solution. Worst performing individuals according to fixed fraction are discarded and new individuals are added to population (Line [11][12][13][14][15][16][17][18][19].…”

Exact and heuristic methods to solve the parallel machine scheduling problem with multi-processor tasks

“…The study also demonstrated the performance guarantee of any list-scheduling algorithm for the P|size j |C max problem should be no less than 2. Switalski and Seredynski (2015) proposed an evolutionary metaheuristic algorithm called the generalized extremal optimization (GEO) to solve a machine scheduling problem with multi-processor tasks. In their study, each machine is defined to have a batch of processors that can work in parallel, and different machines may have different numbers of processors.…”

“…For instance, (1) in diagnosable microprocessor systems, a task must be processed by at least two processors at the same time (Krawczyk and Kubale, 1985); (2) in semiconductor circuit design workforce planning, a design project may need a group of people (Chen and Lee, 1999), (3) in the quay crane scheduling problem of container terminals, a ship requires more than one crane to handle (Trkoullar et al, 2014) and (4) in parallel batch jobs scheduling problems of grid computing environments, a computing task may run on several processors that work in parallel (Switalski and Seredynski, 2015). This paper studies a Parallel Machine Scheduling problem which allows tasks to be processed on more than one machine.…”