Notes on the Implementation of a Remote Fork Mechanism

We describe a method for implementing a remote fork, a primitive with the semantics of a UNIX fork() call which begins the execution of the child process on a remote machine. We begin by examining the subject of process migration, and conclude that most of the relevant process state can be captured and transferred to a remote system without operating system support. We then show how our implementation is easily optimized to achieve a performance improvement of greater than 10 times when measuring execution time. We conclude with some comments on limitations and applications of the remote fork mechanism.