Dynamic Load Distribution Algorithm Performance in Heterogeneous Distributed System

The goal of load balancing is to assign to each node a number of tasks proportional to its performance. Many load balancers have been proposed that deal with applications with homogeneous tasks; but, applications with heterogeneous tasks have proven to be far more complex to handle. Load balancing techniques play a very important role in developing highperformance cluster computing platforms. Many load balancing polices achieve high system performance by increasing the utilization of CPU, memory, or a combination of CPU and memory. However, these load-balancing policies are less effective when the workload comprises of a large number of I/O-intensive tasks and I/O resources exhibit imbalanced load. The I/O intensive tasks running on a heterogeneous cluster needs effective usage of global I/O resources. We have proposed a load-balancing scheme based upon system heterogeneity and migrate I/O-intensive tasks to the fastest processor. The proposed load balancing scheme can minimizes the average slow down of all parallel jobs running on a cluster and reduces the average response time of the jobs.

Distributed heterogeneous computing is being widely applied to a variety of large size computational problems. These computational environments are consists of multiple heterogeneous computing modules, these modules interact with each other to solve the problem. In a Heterogeneous distributed computing system (HDCS), processing loads arrive from many users at random time instants. A proper scheduling policy attempts to assign these loads to available computing nodes so as to complete the processing of all loads in the shortest possible time. The resource manager schedules the processes in a distributed system to make use of the system resources in such a manner that resource usage, response time, network congestion, and scheduling overhead are optimized. There are number of techniques and methodologies for scheduling processes of a distributed system. These are task assignment, load-balancing, load-sharing approaches. Due to heterogeneity of computing nodes, jobs encounter different execution times on different processors. Therefore, research should address scheduling in heterogeneous environment. In task assignment approach, each process submitted by a user for processing is viewed as a collection of related tasks and these tasks are scheduled to suitable nodes so as to improve performance. In load sharing approach simply attempts to conserve the ability of the system to perform work by assuring that no node is idle while processes wait for being processed. In load balancing approach, processes submitted by the users are distributed among the nodes of the system so as to equalize the workload among the nodes at any point of time. Processes might have to be migrated from one machine to another even in the middle of execution to ensure equal workload. Load balancing strategies may be static or dynamic. In static scheduling, the assignment of the tasks to the nodes is done before the execution of the program. Information regarding task execution time and processing resources is assumed to be known at compile time. A task is always executed on the node to which it is assigned. Dynamic scheduling is based on the re-distribution of processes among the processors during execution time. This redistribution is performed by transferring tasks from heavily-loaded processors to lightly-loaded processors with an aim to minimize the processing time of the application. The advantage of dynamic load balancing over static scheduling is that the system need not be aware of run-time behavior of the application before execution. The flexibility inherent in dynamic load balancing allows for adaptation to unforeseen application requirements at run-time. In general, load-balancing algorithms can be broadly categorized as centralized or decentralized, dynamic or static, periodic or non-periodic, and those with thresholds or without thresholds.

Free download research paper