"Partial Replication in the Vesta Software Repository"
Timothy Mann
Report #172, August 17, 2001.

The Vesta repository is a special-purpose replicated file system,

developed as part of the Vesta software configuration management

system.  One of the major goals of Vesta is to make all software

builds reproducible.  To this end, the repository provides an

append-only name space; new names can be inserted, but once a

name exists, its meaning cannot change.  More concretely, all

files and some designated directories are immutable, while the remaining

directories are appendable, allowing new names to be defined but not

allowing existing names to be redefined.



The data stored in a repository can be replicated, to support

distributed software development.  The append-only nature of the

repository greatly simplifies the problem of maintaining consistency

among replicas.  Conceptually, all files and directories stored in all

Vesta repositories are named in one single, global name space.  Each

repository stores some subtree of the complete name space.

Replication is present when the subtrees stored by two different

repositories overlap; that is, some of the same names and data occur

in both.  We call this concept partial replication, because

each repository can choose to replicate all, part, or none of the data

stored in any other repository.



In this paper we outline the main features of the repository,

give a definition for the consistency of partial replicas, describe

how our replication tools maintain consistency, and briefly relate our

experience in using the system for distributed software development

between groups on opposite coasts of the United States.