Name-Based Mapping: Addressing Support for Persistent Objects

Abstract

Systems supporting persistent objects need to address far more information than can fit in an ordinary virtual memory. The usual approach is to impose an additional level of indirection—an object name is mapped to an object number, which serves as part of a virtual address. The virtual address can then be translated in the normal manner. An attractive alternative is name-based mapping, where objects are referred to by name rather than by address. A set of registers and a cache can be used to translate names directly to physical addresses, bypassing the virtual-address middleman. This scheme effectively removes the limit on the size of a process’s address space, since the number of possible names is essentially infinite. Name-based mapping has a number of other advantages. The cache, which cannot be modified by the programmer, is used to translate all addresses. It can therefore be used to enforce different protection domains within a single process. This permits software systems to be built out of larger but fewer processes, decreasing the overhead of managing large numbers of processes. In addition, name-based mapping removes the need for conventional static or dynamic linking. An architecture using name-based mapping can easily be adapted to support object-oriented languages efficiently. Unlike capability-based systems, name-based mapping requires only a special address-translation mechanism, rather than specially protected segments throughout memory. This paper presents the design and initial performance measurements based on simulation.

This research was supported by the Office of Naval Research under contract N00014-88-K-0037.