high performance computing on graphics processing units: hgpu.org

Recently, using GPUs for coprocessing in database systems has been shown to be beneficial. However, information systems processing confidential data cannot benefit from GPU acceleration yet because knowledge of security issues and forensicexaminations on GPUs are still fragmentary. In this paper, we point out key challenges and research questions related to forensics and anti-forensics on GPUs. Our results and discussion are based on analogies from similar computation environments, and experiences. Initial experimental studies indicate that data in GPU RAM is retrievable by other processes. This can be done by creating a memory dump of device memory. Hence, application data is accessible by users without access permissions, by bypassing the access control system of the database management system. Finally, we discuss approaches, how our results can be used in forensic and anti-forensic scenarios.