CUDACS: Securing the Cloud with CUDA-Enabled Secure Virtualization Flavio Lombardi 1 and Roberto Di Pietro 2,3 1 Consiglio Nazionale delle Ricerche, DCSPI Sistemi Informativi, Piazzale Aldo Moro 7, 00185 - Roma, Italy flavio.lombardi@cnr.it 2 Universit`a di Roma Tre, Dipartimento di Matematica, L.go S. Leonardo Murialdo, 1 00149 - Roma, Italy dipietro@mat.uniroma3.it 3 Consiglio Nazionale delle Ricerche, IIT, Via Giuseppe Moruzzi 1, 56124 - Pisa, Italy dipietro@iit.cnr.it Abstract. While on the one hand unresolved security issues pose a barrier to the widespread adoption of cloud computing technologies, on the other hand the computing capabilities of even commodity HW are boosting, in particular thanks to the adoption of *-core technologies. For instance, the Nvidia Compute Unified Device Architecture (CUDA) technology is in- creasingly available on a large part of commodity hardware. In this paper, we show that it is possible to effectively use such a technology to guaran- tee an increased level of security to cloud hosts, services, and finally to the user. Secure virtualization is the key enabling factor. It can protect such resources from attacks. In particular, secure virtualization can pro- vide a framework enabling effective management of the security of possibly large, heterogeneous, CUDA-enabled computing infrastructures (e.g. clus- ters, server farms, and clouds). The contributions of this paper are twofold: first, to investigate the characteristics and security requirements of CUDA- enabled cloud computing nodes; and, second, to provide an architecture for leveraging CUDA hardware resources in a secure virtualization envi- ronment, to improve cloud security without sacrificing CPU performance. A prototype implementation of our proposal and related results support the viability of our proposal. Keywords: Cloud computing security, CUDA, virtualization, trusted platforms and trustworthy systems. 1 Introduction A barrier to the widespread adoption of cloud computing technologies is the num- ber of unresolved security issues. Recent improvements in Graphics Processing Units (GPU) provide the Operating System (OS) with additional computing re- sources that can be used for tasks that are not strictly related with graphics [15]. In particular, commodity hardware such as Ati Stream and Nvidia CUDA [3] fea- ture manycore GPUs capable of efficiently executing most parallel tasks [1]. M. Soriano, S. Qing, and J. L´opez (Eds.): ICICS 2010, LNCS 6476, pp. 92–106, 2010. c  Springer-Verlag Berlin Heidelberg 2010