Automatic Artificial Diversity for Virtual Machines

Large scale adoption of homogeneous computing environments presents serious risk of automated attacks due to the unified nature of the computing environments. Botnet and computer virus attacks are successful due to widespread unification of computing systems, presenting a uniform attack surface so an attack devised for one machine can be replicated to millions of machines. A method available in computing systems not possible for living systems is to change the "DNA" on each individual machine in a cryptographically secure manner, that is, using instruction set randomization. We propose to design and implement an instruction randomization environment suitable for virtual machine deployment. This design will address compiling new code and translating existing binaries to the per machine instruction set, securely selecting instruction sets, implementation issues for the resulting tool chain, virtual machine behavior, and performance issues relating to the interaction of the translated binaries and virtual machine. With this design we will identify diversification opportunities, estimate security gains and possible weaknesses, and detail how the system will function in a production environment BENEFIT: The proposed technology will increase the security of virtual machine platforms, by removing some of the homogeneity through randomization. This randomization makes it harder for automated attacks to have widespread effects such desktops, such as the Federal Desktop Computer Configuration (FDCC). Commercial applications include licensing the technology into current virtualization companies (Microsoft and VMWare) and security companies (McAfee, Cisco, Symantec) as well as offering a product to sell directly to companies needing secure virtualization.