An experimental study of secure outsourcing on scientific computations

Abstract

Although there are many distributed systems or solutions exist in the market for offloading heavy calculation, most of them require expensive cryptographic calculation or key exchange for security rationale. Also, the privacy of user’s data and the outcome of the calculation may be exposed to the untrusted server. As far as we can see, mobile device users want to perform calculation on their devices especially in large volume data processing like stock analysis or algebra calculation. Due to the battery life and hardware limitation, mobile devices are unable to perform such heavy tasks currently. As the growth of mobile device in the market, there are high demands on heavy scientific computation. Securely data exchange between user and distributed system and protecting data privacy of user are also main concerns nowadays; this project presents a solution to facilitate these needs. Our method computes the user data in form of matrices. By matrix properties and linear algebra characteristic, we are able to hide user data from server while man in middle attack can be prevented by this protocol. By taking advantage of GPU computational power and the high speed communication link, our outsourcing calculation will be faster local computation. And this method is secure enough to serve mobile device users in an untrusted linking environment. Finally, the framework of this solution is generic enough to deploy to any powerful server equipped with floating point calculation under different operation system.