The mechanics of in-kernel synchronization for a scalable microkernel

Abstract

Systems with minimal kernels address the problem of ever-increasing system software complexity by strict separation of resource permission management and resource policies into different trust domains. Lately, such system structure has found wide attention in the research community and industry in the form of hypervisors and virtual machines. With an increasing number of processors, these systems face a scalability problem. The separation eliminates semantic information about the expected parallelism for individual resources, such as memory pages or processors. Hence, the kernel is unable to optimize its synchronization primitives on a case-by-case basis - -a precondition for a scalable, yet well-performing system. In this paper we present an adaptive synchronization scheme, one of the core building block for scalable microkernels. Herewith, unprivileged components (like virtual machines) can express the degree of concurrency at the granularity of individual resources. The kernel can safely adapt and optimize its internal synchronization regime on a case-by-case basis as we show exemplary for inter-process communication and the memory management subsystem of an L4 microkernel.