The reputation-based majority-voting approach is a promising solution for detecting malicious workers in a cloud system. However, this approach has a drawback in that it can detect malicious workers only when the number of colluders make up no more than half of all workers. In this paper, we simulate the behavior of a reputation-based method and mathematically analyze its accuracy. Through the analysis, we observe that, regardless of the number of colluders and their collusion probability, if the reputation value of a group is significantly different from those of other groups, it is a completely honest group. Based on the analysis result, we propose a new method for distinguishing honest workers from colluders even when the colluders make up the majority group. The proposed method constructs groups based on their reputations. A group with the significantly highest or lowest reputation value is considered a completely honest group. Otherwise, honest workers are mixed together with colluders in a group. The proposed method accurately identifies honest workers even in a mixed group by comparing each voting result one by one. The results of a security analysis and an experiment show that our method can identify honest workers much more accurately than a traditional reputation-based approach with little additional computational overhead.

Remote data checking (RDC) is a scheme that allows clients to efficiently check the integrity of data stored at an untrusted server using spot-checking. Efforts have been consistently devoted toward improving the efficiency of such RDC schemes because they involve some overhead. In this letter, it is assumed that a probabilistic attack model is adopted, in which an adversary corrupts exposed blocks in the network with a certain probability. An optimal spot-checking ratio that simultaneously guarantees the robustness of the scheme and minimizes the overhead is obtained.