A New Approach to Distributed Memory Management in the Mach Microkernel

In this paper we describe a new approach towards extending Mach virtual memory semantics across the node boundaries of a multicomputer system. At its core, the Advanced Shared Virtual Memory (ASVM) system employs algorithms from the realm of shared virtual memory that were adapted and extended to create a system that is efficient and scalable and supports full VM semantics, paging between node memories and efficient execution of SVM applications. Our performance measurements demonstrate ASVM's superior efficiency and scalability compared to its predecessor, the Extended Memory Manager (XMM) that is part of the Mach kernel's NORMA distribution.