Abstract: Techniques for providing replicated files in a distributed system. A replicated file has a set of copies in components of the distributed system. Operations on the copies have the same semantics for the application processes accessing them as operations on a single copy of a file in a system where all processes execute on the same host. These semantics are achieved by means of a distributed synchronization system. Each replicated file has a read token and a write token. In order for an application process to perform an operation on a replicated file, the token required for the operation must be in the process's host and the process must have access to the token. Tokens are passed between hosts by token servers which respond to requests for tokens from application processes and from other token servers. The techniques are implemented using a library which replaces a standard I/O library, and may thus be employed without modification to hardware or the operating system.