Garbage Collection in Distributed Object Systems

Large-scale, distributed computer systems are less reliable than their smallscale, non-distributed counterparts. A significant proportion of the reliability problems are due to the very poor scalability of manual storage management techniques—the reclamation of unused storage in particular. The bulk of automatic garbage reclamation in both message-passing and shared memory systems, with certain realistic assumptions, can be implemented in a generic, language-independent, low-overhead, fault-tolerant manner that is largely transparent to the programmer. Using such automatic mechanisms reduces the burden of storage management drastically, resulting in applications which are smaller, more reliable, and much easier to understand and maintain.