The new applications that will be offered on the Internet (e.g., electronic commerce) will often be distributed programs providing services based on highly-available access to persistent information. Such systems will be significantly easier to implement if they can make use of a repository that provides highly reliable and highly available storage for persistent objects.

Such a repository not only provides the benefits of object-oriented programming, but in addition can ensure safety in the presence of concurrency and failures by providing support for atomic transactions. Furthermore, the repository can solve several of the hard problems in distributed systems, including consistent cache management and replication protocols for high availability. As a result the application builder can construct a simple sequential system based on the types of persistent objects needed in the application, and can focus on providing sophisticated new user services.

This talk will discuss the requirements for object repositories: what services they need to provide to their users. It will also discuss how to implement object repositories efficiently in a wide-area, large-scale distributed environment.