INRIA Projet SOR ( Systèmes d'objets répartis, or Distributed Object Systems) studies operating system support for information sharing in large-scale distributed systems. The long-term goals of our research are : (i) distributed garbage collection ; (ii) applying distributed systems mechanisms in the large scale. Our targeted application domains are cooperative work in large scale environments (e.g. World-Wide Web) and nomadic computing. We seek system-level solutions, i.e. ones that are general, mutually orthogonal, scalable, application-independent and language-independent.
Our current focus is on the three following issues. First, the persistent distributed shared memory (PDSM) abstraction supports a simple API that programmers are already familiar with. Shared objects are mapped in memory and then accessed via pointers. Persistent objects are those that are accessible from some persistent root. In the past year we proposed Larchant, a PDSM which incorporates a novel garbage collection algorithm. This year we concentrated on three actions : a formal proof of safety and liveness, prototyping Larchant, and designing a post-compiler that extracts type information directly from compiled code.
Second, we provide an efficient and semantically-correct remote reference mechanism. Our Stub-Scion Pair (SSP) Chains are a fault-tolerant reference mechanism for identifying and accessing (possibly mobile and/or replicated) objects remotely. We did a new modular, robust, efficient implementation of the SSP mechanism. Attached to the reference management activity is the implementation of a flexible binding protocol, called Hobbes.
Third, in order to provide programmers with systems that really match their needs, we developped BOAR, a toolbox of components implementing basic sharing supporting mechanisms. BOAR allows application programmers to customize the internal policies of the underlying infrastructure. Our initial targets are fragmented object support, coherence management and storage systems.
Plans for the coming year are to extend our results and their actual application. An exciting field of application is WWW, where our techniques could improve markedly the administrability and level of service perceived by users. We also aim to transfer the Larchant results both to industry and in the public domain.