High performance distributed objects using distributed shared memory and remote method invocation

There are two emerging trends in distributed computing: the evolution of client/server architectures into multi-tiered systems and advances in distributed shared memory (DSM). The convergence of these two trends yields a new structure we call virtual distributed objects (VDOs). The first trend is evolving because of the difficulties involved in programming high-performance client/server applications. Client/server architectures must be designed from the ground-up for good performance. Increasingly, we are seeing new client/server topologies that improve server performance. The second converging trend associated with the transition from client/server to peer-to-peer distributed objects is the rapid advances in techniques for DSM. In the past, DSM has been used primarily for support of scientific applications. While DSM was once envisioned as a means to support an entire paged virtual address space, systems like Mirage and Mirage+ began support for sharing smaller entities called segments that, combined with RPCs (remote procedure calls) and RMIs (remote method invocations), can support VDOs. VDOs allow systems designers to improve system performance by making use of idle machines rather than requiring expensive upgrades to a server or extremely expensive upgrades to all client machines. We describe why high-performance VDOs will become the predominant paradigm for distributed computing in the next decade. We describe why VDOs, supported with DSM and other technologies, could potentially supplant strict client/server structures whose scalability and extensibility are inherently limited