MapReduce is one of the most popular distributed programming frameworks. However, MapReduce in the public cloud suffers from a lack of confidence in the participating virtual machines. Also, malicious nodes may purposely cheat the processing result during map tasks or reduce tasks. Thus, the results will be unreliable and erroneous. In this paper, we propose a technique which overlays on a hybrid cloud. We run the master and some of the slave workers on a private cloud that is a trusted cloud, and the remaining workers run on a public cloud. Our technique depends on replicating a subset of each task to reduce overhead. When a malicious worker on the public cloud executes a task and an error is detected as a part of replicated subset, we detect and exclude this worker from the cloud. We carry out several theoretical experiments to investigate the security and performance overhead. The results provide high computation integrity and little performance overhead.
Atwa, Walid; Abo Aly, Doaa; and Mousa, Hamdy
"Enhancing Map Reduce Computation Integrity on Hybrid Cloud,"
Information Sciences Letters: Vol. 12
, PP -.
Available at: https://digitalcommons.aaru.edu.jo/isl/vol12/iss3/54