Dynamic load balancing of the workloads of clustered workstationshas emerged as a powerful solution for overcoming load imbalance. Inorder to detect such imbalances, some load balancing methods check theaverage idle-time of the workstations periodically. But in these methodsload balancing cannot be performed until the end of a period even ifload imbalance has occurred in the middle of the period. In this paper,we present a new threshold load balancing method for workstations whichprocess the jobs with relatively long execution times. The new methoddecides a proper time to perform load balancing and does perform thebalancing right after the detection of the load imbalance. We also showthat a static load balancing method with a long period is suitable ifthe workstations have to deal with the jobs having unpredictable arrivaltimes and relatively short execution times. The performance of themethods presented in this paper is compared with the method without loadbalancing as well as with the periodic methods in (Siegell andSteenkiste, 1994), (Nedeljkovic and Quinn, 1992) and (Schnekenburger andHuber, 1994). The experiments were done with an iterative data parallelproblem called the ISING problem (Saltz et al., 1995). The experimentalresults show that our methods outperform all the other methods that wecompared
展开▼
