An Efficient Dynamic and Decentralized Load Balancing Technique for Grid

Abstract

Grid computing has recently become one of the most important research topics in the field of computing. The Grid paradigm has gained popularity due to its capability to offer easier access to geographically distributed resources operating across multiple administrative domains. The grid environment is considered as a combination of dynamic, heterogeneous and shared resources in order to provide faster and reliable access to the Grid resources. For efficient resource management in Grid, the resource overloading must be prevented which can be obtained by proper Load Balancing and Job Migration mechanisms. In this scenario, dynamic and decentralized Load Balancing considers all the factors pertaining to the characteristics of the Grid computing environment. Dynamic load-balancing algorithms attempt to use the run-time state information to make more informative decisions in sharing the system load and in decentralization, algorithm is executed by all nodes in the system and the responsibility of Load Balancing is shared among all the nodes in the same pool. For this purpose, in this research work, an extensive survey of the existing Load Balancing and Job Migration techniques has been done. A detailed classification & gap analysis of the existing techniques is presented based on different parameters. A Job Migration and Job Migration approach has been proposed and designed to fulfill all the existing gaps. The issue of Load Balancing in a Grid has been addressed while maintaining the resource utilization and response time for dynamic and decentralized Grid environment. Here, a hierarchical Load Balancing technique has been analyzed based on variable threshold value. The load is divided into different categories, like, lightly loaded, under-lightly loaded, overloaded, and normally loaded. A threshold value, which can be found out using load deviation, is responsible for transferring the task and flow of workload information. In order to improve response time and to increase throughput of the Grid, a random policy has been introduced to reduce the resource allocation capacity etc. Poission process has been used for random job arrival and then load calculation has been done for assigning job to the appropriate Processing Entity (PE) for balancing the load in the pool. After balancing the load, it comes into the normally loaded pool, and then Job Migration process is executed. In Job migration a process running on a Load Balancing resource is redeployed on another one in a way that the migration does not cause any change in the process execution. It means the process is temporarily suspended and later resumed on a new resource. For transferring the job from overloaded node to the underloaded node, the Job Migration system has also been proposed. Here, once the job known as Gridlet is submitted to a resource pool, job queue allocates the PE and starts the Job Migration. If any of the nodes, on which the job processes fail, Local Job Migration Scheduler suspends the job execution and tries to restart it on a new node in the same pool, else switches the job into other resource pool based on the availability of the under loaded node. Grid Information System periodically monitors the status of the job. If it detects that the job has been in the restart state for a long period, it tries to find a new resource pool to which the job can be migrated. Job migration can be achieved by following three procedures. Firstly, Job Migration with check-pointing, Checkpointing needs the small chunks of data to be migrated at the specific time requiring that data to be ready to get migrated. Application instance data on queues and failed events in the source node can be handled through checkpoining process. Next, Job Migration with scheduling, transfers the data on the basis of time and space. At last, Job Migration with replication requires migrating all the applications (such as code, database etc.) at the same time, from the source node to the destination node. So, for this, it needs all the applications on the source node to be ready to get migrated. For above techniques, Job Migration selection policy has been proposed for selecting one of the above techniques. Each of the above techniques has been executed according to its specific condition at run time. To understand Job Migration with replication, a case study Load Balancing and Job Migration in Social Grid Environment_ has been considered further. In the case study, fault tolerance and Quality of Services (QoS) scheduling using Hash Table Functionality in Social Grid Computing has been proposed. Social Grid Computing is a computing model that includes devices to support user mobility. It connects with social networks to reflect real world user relationships, and therefore provides and shares Grid services directly among the members of a social network. Here, Hash Table Functionality (HTF) uses as the underlying Social Grid Computing (SGC) to logically manage the locations of the devices. Fault tolerance and QoS scheduling consist of four sub-scheduling algorithms: malicious-user filtering, Grid service delivery, QoS provisioning, and replication and load-balancing. Under the proposed scheduling, a device is used as a resource for providing Grid services, faults caused by user mobility or other reasons are tolerated and user requirements for QoS are considered. Simulation of scheduling is done, both with and without HTF. The experimental results show that the proposed scheduling algorithm minimizes Grid service execution time, finish time and improves reliability reducing the Grid service error rate. The performance of the proposed model, algorithms and techniques has been rigorously examined over the GridSim simulator using various parameters, such as response time, resource allocation efficiency, etc. Experimental results prove the superiority of the proposed techniques over the existing techniques.