A private name space is a file system name space that is accessible only to a single process and its direct descendants. Private name spaces were originally developed for the Plan 9 operating system. As part of our metacomputing research we have built a private name space system for Unix, v9fs.The current implementation provides a set of servers for both remote file systems and a remote memory-based file system, as well as a "server builders kit" which makes new server construction quite easy. For the clients we have implemented both a virtual file system (VFS) for Linux, and a user-mode library which allows unmodified user programs on (e.g.) Linux, Solaris, FreeBSD, and other Unix operating systems to use the private name space. The system is a `clean-room' implementation of the protocol described on the Plan 9 WEB pages, with a few changes to make it more Unix-compatible, such as support for symbolic links and more efficient directory read operations. The system is distributed under the terms of the GNU Programming License. We use v9fs for both distributed and cluster computing. In a distributed environment private name spaces provide a way to safely access remote file systems without compromising the security of the local or remote systems. In a cluster environment private name spaces allow processes to mount disks from many cluster nodes at once without the reliability risks inherent in NFS cross-mounting. In this paper we describe the private name space implementations as well some of our current uses of private name spaces. 1.0

As part of our metacomputing research we have built a private name space system for Unix. The current system runs completely in user mode, requiring no kernel changes of any kind. The name space can be built by one program and then is inherited via fork by child processes. The system is a `clean-room' implementation of the protocol described on the Plan 9 WEB pages and is `copylefted' and available under the terms of the GNU Programming License. We created this system in order to support our distributed computing environment in a large computer network. Our target environment for this system is a network of 1000