Predictable migration and communication in the Quest-V multikernal

Abstract

Quest-V is a system we have been developing from the ground up, with objectives focusing on safety, predictability and efficiency. It is designed to work on emerging multicore processors with hardware virtualization support. Quest-V is implemented as a ``distributed system on a chip'' and comprises multiple sandbox kernels. Sandbox kernels are isolated from one another in separate regions of physical memory, having access to a subset of processing cores and I/O devices. This partitioning prevents system failures in one sandbox affecting the operation of other sandboxes. Shared memory channels managed by system monitors enable inter-sandbox communication. The distributed nature of Quest-V means each sandbox has a separate physical clock, with all event timings being managed by per-core local timers. Each sandbox is responsible for its own scheduling and I/O management, without requiring intervention of a hypervisor. In this paper, we formulate bounds on inter-sandbox communication in the absence of a global scheduler or global system clock. We also describe how address space migration between sandboxes can be guaranteed without violating service constraints. Experimental results on a working system show the conditions under which Quest-V performs real-time communication and migration.