Dynamic load balancing strategies in heterogeneous distributed system

Abstract

Distributed heterogeneous computing is being widely applied to a variety of large size computational problems. This computational environments are consists of multiple het-
erogeneous computing modules, these modules interact with each other to solve the prob-lem. Dynamic load balancing in distributed computing system is desirable because it is
an important key to establish dependability in a Heterogeneous Distributed Computing Systems (HDCS). Load balancing problem is an optimization problem with exponential solution space. The complexity of dynamic load balancing increases with the size of a HDCS and becomes difficult to solve effectively. The solution to this intractable problem is discussed under different algorithm paradigm.The load submitted to the a HDCS is assumed to be in the form of tasks. Dynamic allocation of n independent tasks to m computing nodes in heterogeneous distributed
computing system can be possible through centralized or decentralized control. In central-ized approach,we have formulated load balancing problem considering task and machine heterogeneity as a linear programming problem to minimize the time by which all task completes the execution in makespan.The load balancing problem in HDCS aims to maintain a balanced allocation of tasks while using the computational resources. The system state changes with time on arrival of tasks from the users. Therefore,heterogeneous distributed system is modeled as an M/M/m queue. The task model is represented either as a consistent or an inconsistent expected time to compute (ETC) matrix. A batch mode heuristic has been used to de-sign dynamic load balancing algorithms for heterogeneous distributed computing systems with four different type of machine heterogeneity. A number of experiments have been conducted to study the performance of load balancing algorithms with three different ar-rival rate for the task. A better performance of the algorithms is observed with increasing of heterogeneity in the HDCS.A new codification scheme suitable to simulated annealing and genetic algorithm has been introduced to design dynamic load balancing algorithms for HDCS. These stochastic iterative load balancing algorithms uses sliding window techniques to select a batch of tasks, and allocate them to the computing nodes in the HDCS. The proposed dynamic genetic algorithm based load balancer has been found to be effective, especially in the case of a large number of tasks.