Abstract
Many grid-computing systems adopt voting-based techniques to resist sabotage. However, these techniques become ineffective in grid systems subject to collusion behavior, where some malicious resources can collectively sabotage a job execution by returning identical wrong results. Spot-checking has been used to detect and tackle the collusive issue by sending randomly chosen resources a certain number of spotter jobs with known correct results to estimate resource credibility based on the returned result. This paper makes original contributions by formulating and solving a new spot-checking optimization problem for grid systems subject to collusion attacks, with the objective to minimize probability of the genuine task failure (PGTF, i.e., the wrong output probability) while meeting an expected overhead constraint. The problem solution contains an optimal combination of task distribution policy parameters, including the number of deployed spotter tasks, the number of resources tested by each spotter task, and the number of resources assigned to perform the genuine task. The optimization procedure encompasses a new iterative method for evaluating system performance metrics of PGTF and expected overhead in terms of the total number of task assignments. Both fixed and uncertain attack parameters are considered. Illustrative examples are provided to demonstrate the proposed optimization problem and solution methodology.