Program analysis for more efficient secure computation
Loading...
Authors
Kennard, Lindsey
Issue Date
2020-12
Type
Electronic thesis
Thesis
Thesis
Language
ENG
Keywords
Computer science
Alternative Title
Abstract
The world of large scale secure computation has changed. Computational needs have far outpaced the ability of most modern companies to build
and maintain their own datacenters. Cloud computation has picked up this
slack and provided very advanced frameworks that allow customers to,
with the click of a button, create relatively low-cost solutions
that can be customized to handle almost any conceivable task.
While cloud computing certainly solves the scalability issue, it introduces
a huge problem: program security.
MPC Amortization is a novel algorithm to better utilize parallelization and amortizationin loops scheduled by MPC compilers. As MPC compilation gains popularity results from traditional compiler research can be leveraged to MPC's specific requirements. Our analysis considers the problem from a program analysis/compilers point of view. It casts the problem in terms of a known NP-hard problem: Shortest Common Supersequence, and presents a scheduling algorithm as well as reasoning about the optimality of schedules. We apply our scheduling algorithm on loops from the literature and present our results.
SecureMCMR utilizes two non-colluding clouds to execute MapReduce programsover encrypted data. One cloud executes the MapReduce program over encrypted data. When it encounters unsupported operations, it sends certain data to the second cloud and the two clouds compute the operation collaboratively. Neither cloud has the ability to view the original input data, and all data that is sent between the clouds are `blinded' hiding their original values. We also present security classification for MapReduce programs and results of running MapReduce applications under SecureMCMR on two clouds: Google and AWS.
Customers often must choose between the flexibility of cloud solutionsand the security of using either self hosted or ``made to order'' secure setups. Still worse, for some customers, especially those working on classified projects, public cloud solutions cannot be used at all. This disconnect has created a demand for a solution that provides security, scalability and cusoimization while still being relatively low cost. In this Thesis we present work on two problems from secure computational. The first, SecureMCMR uses Partially Homomorphic Encryption (PHE) to run MapReduce programs more securely on the cloud. The second, MPC Amortization, is a new scheduling algorithm to improve the performance of loops in Secure Multipartry Computation (MPC) compilers.
MPC Amortization is a novel algorithm to better utilize parallelization and amortizationin loops scheduled by MPC compilers. As MPC compilation gains popularity results from traditional compiler research can be leveraged to MPC's specific requirements. Our analysis considers the problem from a program analysis/compilers point of view. It casts the problem in terms of a known NP-hard problem: Shortest Common Supersequence, and presents a scheduling algorithm as well as reasoning about the optimality of schedules. We apply our scheduling algorithm on loops from the literature and present our results.
SecureMCMR utilizes two non-colluding clouds to execute MapReduce programsover encrypted data. One cloud executes the MapReduce program over encrypted data. When it encounters unsupported operations, it sends certain data to the second cloud and the two clouds compute the operation collaboratively. Neither cloud has the ability to view the original input data, and all data that is sent between the clouds are `blinded' hiding their original values. We also present security classification for MapReduce programs and results of running MapReduce applications under SecureMCMR on two clouds: Google and AWS.
Customers often must choose between the flexibility of cloud solutionsand the security of using either self hosted or ``made to order'' secure setups. Still worse, for some customers, especially those working on classified projects, public cloud solutions cannot be used at all. This disconnect has created a demand for a solution that provides security, scalability and cusoimization while still being relatively low cost. In this Thesis we present work on two problems from secure computational. The first, SecureMCMR uses Partially Homomorphic Encryption (PHE) to run MapReduce programs more securely on the cloud. The second, MPC Amortization, is a new scheduling algorithm to improve the performance of loops in Secure Multipartry Computation (MPC) compilers.
Description
December 2020
School of Science
School of Science
Full Citation
Publisher
Rensselaer Polytechnic Institute, Troy, NY