GeoLoc: Robust Resource Allocation Method for Query Optimization in Data Grid Systems

Overview & Synopsis

Resource allocation (RA) is one of the key stages of distributed query processing in the Data Grid environment. In the last decade were published a number of works in the field that deals with different aspects of the problem. We believe that in those studies authors paid less attention to such important aspects as definition of allocation space and criterion of parallelism degree determination. In this paper we propose a method of RA that extends existing solutions in those two points of interest and resolves the problem in the specific conditions of the large scale heterogeneous environment of Data Grids. Firstly, we propose to use a geographical proximity of nodes to data sources to define the Allocation Space (AS). Secondly, we present the principle of execution time parity between scan and join (build and probe) operations for determination of parallelism degree and for generation of load balanced query execution plans. We conducted an experiment that proved the superiority of our GeoLoc method in terms of response time over the RA method that we chose for the comparison. The present study provides also a brief description of existing methods and their qualitative comparison with respect to proposed method.