Entropy of PRNGs and the Accuracy of Monte-Carlo Simulations in a Publicly Distributed Computing Environment
David J. Hoxie, Saad Raja, Ragib Hasan · 2019
Recently, cloud computing has become of great interest to many computational fields in the physical sciences. Cloud computing offers scalable and economical computational resources to researchers. However, this new paradigm comes at a cost of new security vulnerabilities. While there are methods for ensuring computational attestation for various computations, these methods can be limited, as they may require extensive computation costs, a prior knowledge of boundary conditions or a solution to the computation, or may be limited to deterministic calculations, yet at the very heart of modern physics research lies the requirement for non-deterministic models. Pseudo-random number generators (PRNG) are used for many such experiments to simulate the randomness. It is vital for the PRNGs to have a high degree of entropy. A cloud is an excellent choice for running large scale simulations. However, existing research posit that that PRNGs within virtual machines running in a cloud may possess a lower entropy than that of a local machine. In this paper, we examine this problem. Our results hint that computations in the cloud are equal to that of a local machine and that snap shot vulnerabilities may not apply to these simulations.