Modelling contention sensing memory management systems: A VAX/VMS case study

Digital Equipment Corporation's VAX/VMS operating system maintains a paging cache in memory consisting of modified, free, and shared pages. Pages in the cache are faulted to process resident sets without incurring disk I/O. Pages not in memory are stored on disk and cause a read or 'hard' fault, when referenced. When memory is constrained VMS favors global page reclamation from process resident sets, keeping those processes in memory with smaller resident sets rather than swapping them to disk. This sophisticated scheme to manage paged memory is represented by analytic single class and simulation multiple class models faithful to the details of VMS memory management. Workload dependent equations, developed empirically from experimental data, are used to predict page faults and the percent of which are resolved from disk. Both models were carefully validated against measurement data and predict accurately under different workloads and hardware configurations.