Domain Decomposition vs. Master-Slave in Apparently Homogeneous Systems

Abstract

This paper investigates the utilization of the master-slave (MS) paradigm as an alternative to domain decomposition

Combined load balancing algorithm in distributed computing environment

“…Efficiency of applications using domain distribution algorithm is strongly affected by heterogeneity and variability of distributed computer system. Domain distribution algorithm is the most efficient when the computational problem can be divided into equal parts and computational load is equally distributed among the processor cores [13]. Dynamic load balancing is based on the rescheduling of processes among the CPU cores during execution time of parallel program.…”

“…In addition, many modern supercomputer architectures (such as multicore or SMP clusters) that look homogeneous from the outside actually conceal a heterogeneous and dynamic environment on the inside. For instance, processors located within the same node are actually competing for shared resources, and intra-node communication is typically much faster than inter-node communication [13]. Losses during execution of parallel applications can happen in imbalanced applications as a result of CPU cores idle time once they have finished their work and wait for slowest core or group of core to finish the given tasks.…”