Initial file-placement in data grid environment using game theory and fictitious play
Loading...
Date
2009
Authors
Indrayanto, Aloysius
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Data allocation comprises data placement or migration. It is one of the main considerations
in the design of a distributed data sharing system. The use of equations which only depends on
mostly static input parameters would cause only the best data servers to be selected. It would
eventually overload the selected data servers.
This thesis proposes an algorithm for optimizing initial file-placement in Data Grid Environment
(DGE). The proposed algorithm considers the content of the files and the condition of
the system as its basic parameters. Fictitious play, a belief-based learning rule, is adapted to
help predicting the overall placement behavior of the system. Game theory, a theory of conflict
resolution, is applied on top of the belief and the system’s file-transfer history to provide load
balancing.
Due to the nature of the Grid system and the limitation of our testbed, it was decided to test
the proposed algorithm using simulation. Therefore, a sequential (serial), discrete-time based
simulator was developed. From the simulation results, the use of both belief and history generally
achieved better query performance in most of the simulation groups with 0:443 seconds
(0:137%) to 3:188 seconds (2:067%) faster than using none. However, the use of only history
was able to achieve the best performance with 5:462 seconds (3:161%) faster than using
none. Meanwhile, the use of only belief achieved the worst performance in one of the simulations with 6:976 seconds (4:344%) slower than using none. These results are due to how the
algorithm observes and predicts the overall system behavior.
Description
Master
Keywords
Computer Science , Data grid , Game theory , Fictitious play