1. All types of HEP data (raw, reconstructed, summary and tag) should be accessible as Objects and Containers (collections of Objects) in a uniform manner.
2. It should be possible to make dynamic selections of events and associated event information within the federation of Object Databases. The Object Query Language (OQL) used to make the selections should be sufficiently powerful, intuitive, and customizable for individual and small groups of users.
3. The system should optimise access bandwidth according to data access patterns, by moving frequently accessed data to higher performance media, and migrating infrequently used data lower in the storage hierarchy. This part of the evaluation will test both the Object Database Management System (ODBMS) and its coupling to the HPSS. The system should also determine relations between the stored objects based on patterns of how they are accessed, and then "cluster" the objects accordingly. Additionally, it should be possible from the applications code to add information on the relations between collections of objects.
4. The system should be capable of providing access to PetaBytes of data. In the early stage of the project, a more modest amount of data will populate the database. The size should exceed that being used in all other ongoing projects which will cover the range up to 1 TByte.
5. The system must use information on the network performance, configuration and prevailing conditions to evaluate a cost function for accessing objects stored at remote centres. An initial systems-simulation model to help predict and track the network-related performance issues, ModNet, is described on a separate sheet.
6. The system must allow access to the data by several applications and users simultaneously. The locking system must be robust and efficient, and not have a significant impact on the overall performance of the system.
7. In order to meet the above requirements, the system must exhibit an efficient hierarchy of caching across several levels: single processor/node/cluster/shared memory/disk/tertiary storage, with extensions across very high speed LANs and high speed WANs.