Cooperation of Mutually Suspicious Subsystems in a Computer Utility

Mike Schroeder

This thesis describes practical protection mechanisms that allow mutually suspicious subsystems to cooperate in a single computation and still be protected from one another. The mechanisms are based on the division of a computation into independent domains of access privilege, each of which may encapsulate a protected subsystem. The central component of the mechanisms is a hardware processor that automatically enforces the access constraints associated with a multidomain computation implemented as a single execution point in a segmented virtual memory. This processor allows a standard interprocedure call with arguments to change the domain of execution of the computation. Arguments are automatically communicated on cross-domain calls – even between domains that normally have no access capabilities in common. The processor, when supported by a suitable software system which is also discussed, provides the protection basis for a computer utility in which users may encapsulate independently compiled programs and associated data bases as protected subsystems, and then, without compromising the protection of the individual subsystems, combine protected subsystems of different users to perform various computations.