There exist many mechanisms, ranging from processes to virtual machines, for isolating untrusted computations from each other. Each mechanism explicitly isolates certain resources while, either implicitly or explicitly, sharing the rest. Unfortunately, we lack a comprehensive way to formally and systematically reason about which resources are shared, to what extent they are shared, and how this sharing determines the degree of isolation between any two computations.

DRAM now accounts for over 30% of overall datacenter expense [30], due to its increasing cost and decreasing scaling. [19, 22]. As applications demand more memory, operators look for cost-effective solutions to handle these increasing requirements.