Modeling and performance evaluation of colluding attack in volunteer computing systems

This paper proposes a model of a colluding attack in Volunteer Computing (VC), where some of participants are unreliable and may return incorrect results. The colluding attack is one of the most important issues to realize reliable VC because it may ruin the most basic assumption, i.e. "the majority in voting is correct". Especially, check-by-voting archives the largest efficiency by sorting reliable participants based on whether their results are the majority or not. Thus, if some incorrect results become the majority by colluding attack, it may have significant impacts on check-by-voting. In this paper, we perform a Monte Carlo simulation of VC using the proposed colluding model and evaluate the sabotage-tolerance performance of voting methods. Simulation results show that check-by-voting works well if colluding attack happens.