Information storage in a decentralized computer system
David Kenneth Gifford · CERN Document Server (European Organization for Nuclear Research) · 1981
This paper describes an architecture for shared information storage in a decentralized computer system. The issues that are addressed include: naming of files and other objects (naming), reliable storage of data (stable storage), coordinated access to shared storage (transactional storage), location of objects (location), use of multiple copies to increase performance, reliability and availability (replication), dynamic modification of object representations (reconfiguration), and storage security and authentication (protection). A complete model of the architecture is presented, which describes the interface to the facilities provided, and describes in detail the proposed mechanisms for implementing them. The model presents new approaches to naming, location, replication, reconfiguration, and protection. To verify the model, three prototypes were constructed, and experience with these prototypes is discussed. The model names objects with variable length byte arrays called references. References may contain location information, protection guards, cryptographic keys, and other references. In addition, references can be made indirect to delay their binding to a specific object or location. The replication mechanism is based on assigning votes to each copy of a replicated object. The characteristics of a replicated object can be chosen from a range of possibilities by appropriately choosing its voting configuration. Temporary copies can be easily implemented by introducing copies with no votes. The reconfiguration mechanism allows the storage that is used to implement an object to change while the system is operating. A client need not be aware that an object has been reconfigured. The protection mechanism is based on the idea of sealing an object with a key. Sealed objects can only be unsealed with an appropriate set of keys. Complex protection structures can be created by using such operators as Key-Or and Key-And. The protection mechanism can be employed to create popular protection policies such as capabilities, access control lists, and information flow control.