Distributed System

The Sun Ray platform of the late 90s was an impressive system, ahead of its time. Start an X Window session on one terminal, yank your smartcard, put it into another terminal, and your original session restarted. It was no longer running on the original system and doing a remote protocol, the session was transferred to the new system. Jump ahead to current day, where home servers for consumers, and the thin-client model for corporations, are becoming popular again. Ease of administration, less components to break / cheaper systems to add, etc.

Couple the resurgence of a thin client with the growing movement behind grid computing, and a properly designed distributed OS such as Ensemble can position itself to provide the benefits of ease of administration in thin clients, the ubiquity of processing power and storage space of grid computing, and the safety and control of local systems, all in one platform.

For the average home user, there are two options:

• Purchase not-so-thin clients that have some processing power to them, connected via a normal broadband connection to an Ensemble-based Environment Service Provider computing grid. Apps would be served up from the ESP, as well as user data. The user can also elect to have portions or all of their data stored on the client system's local storage. The relieves the average home user from having to administer their system (patches, backups, etc), while also providing some measure of optional control over the ownership and privacy of their data (one of the main arguments against hosted / grid computing).
• Purchase a reasonably powerful central home server, which handles most computing and storage for thin clients in the home. This again removes the need to patch multiple systems, relieves a home user with many systems from remembering what data is on what system under what shared drive, etc. The central home server can optionally distribute to an aforementioned Environment Service Provider grid, when additional computing or storage power is temporarily necessary.

For a small to medium corporate environment, the same general options exist:

• A tiny office can run dedicated off of an ESP grid, or run a central office computing server.
• A larger office can connect together multiple compute servers into a local office grid.

A large enterprise can tier compute servers, partition off computing grids, and isolate system availability by departments, divisions, or however else their corporate policies dictate. For more computing-intensive situations, tasks can be distributed across the compute grid, the excess power left in the not-so-thin clients, and also from external ESP grids if so elected.

Ensemble should be designed to present a unified computing cloud, optionally showing local storage for more sensitive data. As more clients are needed, one need only purchase simple thin clients. More advanced users can be given either thicker clients or full-blown dedicated PCs. But the heart of the system is the grid of one or more compute servers, combining the administration advantages of thin-client computing with the easy scalability of grid computing.