Article Preview
TopIntroduction
In grid computing system, a large number of geographically distributed resources collaborate to serve the computational need of the users. Due to the limited computing capability of desktop machines, it is often impossible to execute a complex and compute intensive application that requires huge computational power on such machines. A grid, a collaborative distributed system, facilitates such complex computations. Grid resources, normally heterogeneous and geographically distributed, allow parallel job execution and also increase the resource usage efficiency of the system considerably.
Grid resources are normally under the purview of different administrative domains and thus define varying usage policies. These grid resources may even have different capabilities and configurations, making grid resource management a core and important issue for a viable grid system (Raza & Vidyarthi, 2008; Foster, 2002). Resource management in a grid warrants the efficient allocation of jobs to the grid resources which makes job scheduling a critical task in a grid environment (Foster, 2002; Izakian, Ladani, Zamanifar, Abraham & Snasel, 2009; Sharma, Soni, Mishra & Bhuyan, 2010).
Energy efficient resource usage or green computing is attracting the attention of researchers across the discipline as an important aspect. Grid computing also consumes huge amount of electrical energy for managing the grid resources so is a natural choice for energy optimization. Though, the grid system heterogeneity and the users’ constraints make resource allocation in computational grid a difficult problem, the problem gets further aggravated when energy optimization is also considered along with other QoS performance measures.
Though the objective of the computational grid is to utilize the resources of the system to its fullest extent, still not all the grid resources are used all the time. Often, many of these devices remain unused for a variable period, wasting not only the huge computing warehouse but also resulting in electric energy wastage. The grid resources, which may not contribute to an execution, still consume energy. Such energy wastage can be saved. By virtue of the selective connectivity, it is possible to accomplish the user’s task satisfying the desired constraints and putting the idle i.e. no load nodes into sleep mode, thus conserving the electrical energy (Develder, Pickavet, Dhoedt & Demeester, 2008; Narsingh, 2004; Orgerie, Lefèvre & Gelas, 2008; Ponciano & Brasileiro, 2010).
The idea of resource allocation problems, that normally aims to obtain minimal execution time for the job execution by effectively managing the available grid resources, may be extended for minimal energy consumption while meeting the users' and systems' constraints. Grid houses a large number of resources to facilitate high performance computing at the cost of huge electrical energy consumption. It is important that the grid system, not only should meet the desired Quality of Service but also be energy efficient.
The proposed model, focusing on green energy, is designed to suit best the conflict of interests of both; the system and the users. Once a job is submitted, the first part of the proposed model identifies the best grid cluster based on the user’s requirements and available system resources. Best available resources were explored based on the system load at that time and the user’s requirements, i.e. in terms of job specialization, degree of concurrency or the turnaround time expected/offered. In the second part, the energy saving mechanism is implemented that uses a dynamic threshold method followed by a graph theory concept (Narsingh, 2004) i.e. MST (Minimum Spanning Tree).
The dynamic threshold method suggests that all the nodes with the load less than a dynamic evaluated threshold are put in the sleep mode. The load of the sleeping nodes is redistributed among other active nodes while maintaining the execution time constraint. MST algorithm is then used to reduce the number of active links by restructuring the grid. The MST through consolidation maintains an acceptable performance level in the job execution. The selective connectivity, using this algorithm, ensures a further reduction in the overall energy consumption, but maintains the completion time of the job (Abdulgafer, Marimuthu, & Habib, 2010; Xu, Shang, Li & Wang, 2013).