A Load balancing algorithm based on parallel computing entropy in HPC
Yang Xun · JOURNAL OF SHENZHEN UNIVERSITY SCIENCE AND ENGINEERING · 2007
The parallel computing entropy (PCE) was defined to measure the load equilibrium in Beowulf systems and the rule of PCE maximization was proposed. On this basis, a novel algorithm was presented to balance the loads among computing nodes. Task amount instead of task number was employed to measure the load dispatched to nodes in the algorithm. A load was transferred from its original node to another if such a transfer could increase the PCE. The association between the entropy and the program executing time was evaluated. The algorithm was applied to the multiplication of large scale matrices as an example to show its effectiveness. The experimental results indicate that the PCE and executing time are highly associated and the algorithm is effective.