Distributed Shared Virtual Memory for Interoperability

A “subtle shift” in the underlying computer hardware architecture has emerged recently. Although this shift rises from the natural tendency to expand to larger systems with greater capacities, the significance of the step has not been fully exploited by computer users and experts per se. The “shift” is the movement from the 32-bit address architecture to the 64-bit one (called wide-address space architecture). Only a minority from the research and industrial communities have recognized the potential of this change. Human nature keeps many fettered in the familiarity of the past with a preference on evolutionary rather than revolutionary change. The shift to 64-bit architectures (and beyond) has the potential to revolutionize the way information is managed and shared. The reason is that 64-bits provide a very large address space that can be effectively utilized as a single shared memory between multiple autonomous applications. Building on this philosophical faction, we propose a persistent distributed shared virtual memory (DSVM) for distributed object management (DOM) and interoperability of information systems. The DSVM provides a single persistent shared virtual address space (SVAS) that is common to all inter-operating systems across all sites. An object is created in the SVAS at a particular virtual address, which serves as its immutable object identity (oid) throughout its existence. The DSVM ensures that objects are consistent and that they appear at the same location over all systems. The actual storage and management of objects by the DSVM is distributed over multiple sites. The pressing question addressed in this paper is: Can DSVM effectively support interoperability? The systems that should be able to interoperate include legacy applications (e.g., flat files, spreadsheets), traditional database management system (DBMS) platforms (e.g., network, relational), and non-traditional platforms such as object-oriented DBMSs, GIS, MIS, and CAD/CAM, to name a few. These systems are characterized as arbitrarily complex, autonomous, and may desire to interact with one another in fairly sophisticated ways. We argue that existing technology can support the construction of a DSVM system for this type of interoperability environment. A fully interoperable system is one that meets the transparency guidelines imposed by homogeneous distributed systems while maintaining the autonomy of each participating system. Any multi-operable system that meets this criteria will be very useful. Therefore, the ideal interoperable management system architecture (IMSA) must support the autonomy of participating systems and the transparencies outlined in [8]: data independence, network transparency, replication transparency, fragmentation transparency, and language transparency. Unfortunately, supporting each transparency can only be accomplished with current approaches when the autonomy of the underlying local (or participating) DBMS is sacrificed or violated to a great extent. However, by exploiting DSVM built on a suitable architecture, such as the wide-address space model, each of these transparencies can be supported without undue autonomy violations. Due to space restrictions, it is impossible to detail precisely how each transparency is supported by the DSVM model. However, this paper contributes by addressing three important aspects that demonstrate the utility of the approach. Specific attention is paid to each of the following questions: