Building Secure Resources to Ensure Safe Computations in Distributed and Potentially Corrupted Environments

Abstract Security and fault-tolerance is a big issue for intensive parallel computing in pervasive environments with hardware errors or malicious acts that may alter the result. In [10, 15]is presented a novel, robust and secure architecture able to offer intensive parallel computing in environments where resources may be corrupted. Some efficient result-checking mechanisms are used to certify the results of an execution. The architecture is based on a limited number of safe resources that host the checkpoint server (used to store the graph) and the verifiers able to securely re-execute piece of tasks in a trusted way. We extend this approach in the case we have also some secure processors to build a trusted check-pointing infrastructure and to replace some untrusted nodes, avoiding some re-execution. Our approach is illustrated on a medical application and some experimental results are presented