A Dynamic Scheduling Algorithm in Heterogeneous Computing Environments

Abstract

the grid computing system can support execution of computationally intensive parallel and distributed applications. The main characteristics of grid computing system are similar to the heterogeneous computing system, except the fluctuant transfer rate. A dynamic scheduling algorithm, called Dynamic HEFT (DH) algorithm is proposed in this paper to enhance the functions of the original static HEFT (SH) algorithm. Instead of dispatching tasks to physical processors directly, the DH algorithm dispatches tasks to multiple queues. During runtime, the DH algorithm continues to dispatch the scheduled tasks to corresponding physical processor and predicts the transfer rate. Once the difference between two consecutive transfer rates is greater than the threshold value, rescheduling will be performed. The experimental results show that the proposed DH algorithm performs better than the SH algorithm in the system, especially under the grid computing environment, with fluctuant transfer rate and high bandwidth differences.

A Level-wise Priority Based Task Scheduling for Heterogeneous Systems

“…The performance of the algorithm is compared with respect to different number of data points, which varies from 2 to 16 incrementing powers of 2, and the result is shown in Fig.6. Further the efficiency of the algorithm is compared for various number of processors (2,4,8,16,32) and is shown in Fig.7 and Fig.8. Fig.…”

“…The average communication cost of sending data from task t i to task t j is defined by (2) where is the average communication startup costs over all processors, is the average communication cost per transferred unit over all processors. If t i and t j are on the same processor then = 0 since intraprocessor communication is negligible.…”

“…In static scheduling, decisions can be made at compile-time, since the application characteristics such as execution time of tasks, data dependencies between the tasks and the amount of data to be transferred between the tasks are known priori, whereas in the dynamic scheduling decisions are made at run time [2]. The static task scheduling for a heterogeneous distributed computing system is an NP-complete problem [1], which means that there is no known algorithm that finds the optimal solution in polynomial time.…”