Implementation and evaluation of credibility-based voting for volunteer computing systems

Abstract

This paper presents the performance of credibility-based voting in a real volunteer computing (VC) system. In VC systems, a management node distributes a computation to multiple participants, collects the result candidates, and then performs a voting to eliminate incorrect results. In this paper, we focus on the credibility-based voting as the voting method, which is known to be an efficient method to balance the redundancy and reliability of the computations. Although theoretical performance of the credibility-based voting has already studied and is shown to be better than a popular voting method, its real performance has not yet been evaluated. We implement the credibility-based voting in a multi-threaded management node in a real VC system and evaluate the performance varying the system scale. From the experimental results, we show that, in contrast to the theoretical study, performance degradation is observed and is proportional to the number of participants. We also show that the performance degradation comes from the overhead of database I/O in the management node.