Accommodating RPC Heterogeneities In Large Heterogeneous Distributed Environments1

Many RPC semantics have been designed and implemented in recent years to meet various applicationspecific requirements. Examples are synchronous RPC, asynchronous RPC, fault tolerant RPC, broadcast RPC, maybe RPC (no-return RPC), RPC with atomic transactions, and RPC with call-back mechanism [I, 21. With emerging applications like multimedia conferencing and distributed real-time applications, it is conceivable that even more RPC protocols will be designed and implemented. This diversity of RPC protocols makes us adopt a general view of RPC as a protocol above OS1 transport layer in this paper. The problem with having many different RPC protocols is that user programs built on top of different RPC protocols cannot be interconnected directly, greatly reducing the availability of software and resources in a large heterogeneous distributed environment. This problem has been addressed by HRPC/HCS [3] on a smaller scale, where the number of different RPC instances is small, RPC protocols are similar, and new RPC instances are rarely introduced. However, these characteristics do not always hold in a large heterogeneous environment. In such an environment, an acceptable solution must handle a large number of RPC instances, diversified RPC protocols, and rapid RPC protocol evolution at low software development and maintenance costs. Our RPC agent synthesis scheme is a such solution [4]. Our system also includes a mechanism for distributing and acquiring synthesis information , and it supports RPC protocol evolution with minimumdisturbance to the environment. Our approach is best for cross RPC within the same class of RPC semantics, for example, cross RPC among at-most-once RPCs, or among at-least-one RPCs. Although cross RPC between dissimilar RPC classes is also allowed, it is