Article Preview
TopIntroduction
Computational grid (Foster & Kesselman, 1998) provides us a powerful ability to solving the large sale problem efficiently in modern distributed computing systems. Experiments on the grid computing can usually be performed on the hardware framework or the software simulation. The framework of computational grid can be established by the middle-wares, such as Globus Toolkit (Globus, 2011), Oracle Grid Engine Software (Oracle, 2011), and gLite (2009), etc. We know that the hardware framework is more realistic than software simulation. However, high costs for devices and complex system management result in significant difficulties when grid researchers intend to build a large-scale grid environment. Comparing with building a large sale hardware platform, software simulation is more elastic and economic for researchers at the beginnings of analyzing the grid system.
Related simulation tools for grid computing are MicroGrid (Liu, Xia, & Chien, 2004; Song, Liu, Jakobsen, Bhagwan, Zhang, Taura, & Chien, 2000), GridSim (Buyya & Murshed, 2002), and GridG (Lu & Dinda, 2003), etc.; and the grid resource models can be found in these simulators. For simulation tools, the grid resource models have direct influence on simulation result. The common drawback of these grid resource models is usually simple and not realistic enough, so we need some real evidences to make up its completeness, and improving the accuracy of software simulation becomes a critical issue. In our previous work, an Internet-scale grid resource topology generator with aggregative characteristics has been discussed and established (Chen, Chen, Wang, & Lin, 2007). The grid resource topology can be easy generated and integrated with existing grid simulators to perform numerous experiments. Moreover, by using the topology transformation technologies, such as, the PaGrid (Huang, Aubanel & Bhavsar, 2006) or a proposed virtual mesh transformation method on a computational grid (Chen & Lin, 2010), the irregular grid resource topology can be transformed into a virtual mesh. The benefit is that researchers can construct a problem-solving environment on a computational grid so as to solve a large-scale mesh-based problem by traditional parallel processing algorithms (Grama, Gupta, Karypis, Kumar, 2003). The mesh-based problems including matrix multiplication, sorting, fast Fourier transform (FFT) and partial differential equations (PDE), etc., are of fundamental importance to the scientific computing applications (Karniadakis & Kirby, 2003). Before performing these topology transformations, the realistic grid resource topology surely takes a critical position in the experimental simulation. This research puts emphasis on how to build a grid resource topology with realistic characteristics. To carry on the progress of this topic, the characteristic distribution of real grid resource topology such as computational ability and communicational ability will be analyzed and discussed in this paper. The existing grid simulators just simply or randomly configure the resource models without consideration of characteristics in the real grid resources. This effort is unprecedented and is made to rectify this problem.
The remainder of this paper is organized as follows. The next section defines the grid resource model and its essential components. The proposed approximating method is introduced step by step in Section Methodology, and it is illustrated by two examples. In the fourth section, the experiment results are presented with a discussion. The final section draws a conclusion for this research.