MHGrid: Towards an Ideal Optimization Environment for Global Optimization Problems using Grid Computing M. Wahib Hokkaido University Graduate School of Information Science N 14 W 9, Kita-ku Sapporo 060-0814, Japan wahibium@uva.cims.hokudai.ac.jp Asim Munawar Hokkaido University Graduate School of Information Science N 14 W 9, Kita-ku Sapporo 060-0814, Japan asim@uva.cims.hokudai.ac.jp Masaharu Munetomo Hokkaido University Information Initiative Center N 11, W 5, Kita-ku Sapporo 060-0811, Japan munetomo@iic.hokudai.ac.jp Kiyoshi Akama Hokkaido University Information Initiative Center N 11, W 5, Kita-ku Sapporo 060-0811, Japan akama@iic.hokudai.ac.jp 1. Introduction This paper introduces MHGrid, a framework that ex- ploits meta-heuristics based search methods and Grid com- puting to enable the transparent sharing of heterogeneous and dynamic resources offering a Grid based Global opti- mization framework. MHGrid allows a user to solve almost all kinds of global optimization problems in a Black Box manner with a minimal input from the user, it also allows the user to integrate his own solver into MHGrid. In this pa- per we will discuss the architecture and motivation of such a system. We will also discuss the challenges/complexities involved in constructing MHGrid. 2. Motivation for building MHGrid Not many projects so far have targeted for a general pur- pose optimization environment over Grid. Nimrod/O[1] and GEODISE[2] are mature projects in this domain, but they still are both targeting special purpose problems (fluid dy- namics in case of GEODISE for example), they don’t al- low the user to add his own solver like NEOS[3] and offer solutions as problem solvers not as services. MHGrid as an optimization solving environment should be distributed due to the very high potential of parallelization in the Meta- heuristics [8]. Building a universal optimization framework based on the Grid technology was motivated by the follow- ing: 1. Parallel implementation on Grid offers greater advan- tages over conventional parallel computing methods like cluster or super computing because of the ability to exploit geographically dispersed heterogeneous re- sources, for example the Grid can help if you have a fitness function that involves taking a reading from a telescope and communication with a satellite, in this case conventional distributed computing like cluster computing will be of a little help. And for our case in MHGrid, the resources include clusters at Informa- tion Initiative Center, Hokkaido University as well as other clusters with a distance of over 1200 Km with heterogeneous resources. 2. The need to build a framework following standardized technologies and tools, to assure the ease of interoper- ability with other Grid systems and to allow extendibil- ity and categorization. 3. The variable requirements of the users seeking opti- mization solving. Traditionally, the major concern of a user trying to solve his optimization problem was time and accuracy. As the Nature of the Large Scale networking and high performance computing changed dramatically, now a user may have other concerns when approaching a solution for his optimization prob- lem. Objectives include the demand of locally unavail- able dataset, execution of an expensive computational problem regardless of the round time. 3. Architecture of MHGrid Figure 1 gives an overview of the MHGrid Architecture. The base layer is a high performance Grid network. On Eighth International Conference on Parallel and Distributed Computing, Applications and Technologies 0-7695-3049-4/07 $25.00 © 2007 IEEE DOI 10.1109/.62 167 Eighth International Conference on Parallel and Distributed Computing, Applications and Technologies 0-7695-3049-4/07 $25.00 © 2007 IEEE DOI 10.1109/.62 167