Adaptive bidding load balancing algorithms in heterogeneous distributed systems

Abstract

Describes three adaptive bidding load balancing algorithms, each of which balances the workload on hosts (nodes) using different amounts of system state information in a heterogeneous distributed system model. A key feature of these algorithms is that they attempt to balance the virtual delay (unfinished work) of each node. Using simulation, the algorithms are examined and compared with Eager's (1986) 'shortest algorithm', an efficient algorithm in homogeneous systems. Results show that the simple algorithms we consider in this paper, which use very little system state information, yield dramatic performance improvements over the case of no load balancing and perform much better than Eager's algorithm.<>