Foreword

In a secure outsourcing system, the clients would like to perform operations that are computationally complex while their local machines do not have enough resources to solve the problem. With the emerging cloud technology, the clients could easily obtain computational resources they needed from cloud hosts. Currently, the major cloud hosts include Google Cloud Platform (GCP), Amazon Web Services (AWS), and Microsoft Azure. However, the problem might contain sensitive information that the clients would not want to disclose and computed outside of a controlled environment. This imposes a challenge: how to securely outsource computationally intensive computations to cloud servers without expose sensitive information.

Currently, there are existing algorithms that resolve the problem to limited scopes. Some of them relies on specific mathematical properties of the input problem, while some others requires a large amount of computational resources on local. Most importantly, almost all existing method imposes a lot of memory I/O operations on local machine. The resource limitations on local machines has been barely addressed, as in most of the cases, local machines lacks enough random access memory (RAM) to load the problem. Our algorithm fixed the problem by only perform necessary secure transformation locally to keep the task simple for the client machine while outsource computationally complex operations to the cloud.

In this project, we implemented the algorithms using the proposed schemes. That is, the cloud cooperates with the local machine to compute the result without compromising the privacy and security. To perform the outsourcing, we used Dask, a distributed parallel computing platform and library 1. Users may setup their Dask server in a potentially unsafe environment and is able to solve their problems securely using our algorithms.

References

1

  1. Rocklin, “Dask: Parallel Computation with Blocked algorithms and Task Scheduling,” presented at the Python in Science Conference, Austin, Texas, 2015, pp. 126–132 [Online]. Available: https://conference.scipy.org/proceedings/scipy2015/matthew_rocklin.html. [Accessed: 14-Jul-2019]