Modelling User Behavior and Requirements for Job Scheduling in Computational Grids

Abstract

Computational Grids (CGs) are a type of distributed system that virtually combine geographically distributed IT resources from many different administrative domains into one single customized computational infrastructure, CGs enable users to perform computational tasks or data storage capabilities in a transparent and secure manner. Unlike traditional distributed systems belonging to single administrative domain and having a few user types, in CGs several user types should co-exist and make use of resources according to the hierarchical nature and the presence of the multiple administrative domains, which impose different access and usage policies on resources. In the talk, we firstly highlight the most common Grid users types and their relationships and access scenarios in CGs corresponding to traditional requirements in Grid scheduling such as performance, and new requirements such as security and trust. We identify and analyze new features appearing in users’ behavior in Grid scheduling, such as dynamic, selfish, cooperative, trustful, symmetric and asymmetric behavior. Analyzing and modelling such user requirements and behaviors to predict the users needs and actions are important in order to optimize the Grid system performance at individual and global levels. Game theory in combination with economic theory is playing an important role in Internet computing to develop algorithms for finding equilibria in computational markets, computational auctions, Grid and P2P systems as well as security and information markets. The use of game-theoretic modelling of user behaviors in scheduling and resource allocation in CGs enables a highly scalable and efficient decision-making processes. We highlight the advantages and limitations of non-cooperative, cooperative and semi-cooperative game models based on assumptions that game players are rational and pursue well-defined objectives and they take into account their knowledge or expectations of other players behavior. Artificial Neural Network (ANN) is another useful approach for supporting new user requirements such as security awareness. Making a prior analysis of trust levels of the resources and security demand parameters of tasks, the neural network is monitoring the scheduling and task execution processes. The network learns patterns in input (initial tasks and machines characteristics) and produce the tasks-machines mapping suggestions as the outputs. Finally, while game-theoretic and ANN approaches are useful at modelling Grid users behaviors and requirements, they are not effective as stand alone approaches for solving the multi-objective optimization problems arising in such models. We then show how game-theoretic and neural networks can be combined with meta-heuristic approaches, such as Genetic Algorithms, to solve the optimization problem and achieve Grid system performance.