Offloading an encrypted user space file system on Graphical Processing Units

Abstract

Modern computers often have a powerful graphics processing unit (GPU), either on dedicated graphic cards or integrated on the motherboard. These units can used by applications for demanding computation. Another use of this technology is to assist the main CPU in the system, offloading some of its work. Offloading can give increased performance, as well as decreased load on the main CPU. Decreasing the load frees resources for other applications.

Keeping documents, images and other potentially sensitive files private is important for many users. One way to do this is to use an encrypted file system, which can pre- vent others from gaining unauthorized access. However, such a file system occupies resources in the computer system. In this thesis, we evaluate how GPUs can be used for assisting the computationally expensive encryption part of an encrypted file system.

Programming GPUs, is challenging because of the GPUs massively parallel nature and their many memory types. We will look into different architectures, focusing mainly on NVIDIAs architecture and programming framework in our work on evaluating the effects of using graphic processing units for offloading an encrypted file system. In this thesis, we see that offloading parts of this file system is beneficial, giving better performance and reduced CPU load.