ROBUST CONCURRENCY CONTROL FOR A DISTRIBUTED INFORMATION SYSTEM

• Authors: W. A. Montgomery
• Venue: ROBUST CONCURRENCY CONTROL FOR A DISTRIBUTED INFORMATION SYSTEM
• Year: 1979

This dissertation presents a collection of protocols for coordinating transactions in a distributed information system. The system is modeled as a collection of processes that communicate only through message passing. Each process manages some portion of the data base, and several processes may cooperate in performing a single transaction. The thesis presents a model for computation in a distributed information system in which the sites and communication links may fail. The effects of such failures on the computation are described in the model. The thesis discusses implementation techniques that could be used to limit the effects of failures in a real system to those described in the model. A hierarchical protocol for coordinating transaction are pre-analyzed to select the protocols needed to coordinate the processes that participate in the implementation of the transaction. This analysis can be used to guide the organization of the data base so as to minimize the amount of locking required in performing frequent or important transactions. An important aspect of this mechanism is that it allows transactions that cannot accurately be pre-analyzed to be performed and correctly synchronized without severely degrading the performance of the system in performing more predictable transactions. A novel approach to the problem of making updates at several different sites atomically is also discussed. This approach is based on the notion of polyvalue, which is used to represent two or more possible values for a single data item. A polyvalue is created for an item involved in an update that has been delayed due to a failure. By assigning a polyvalue to such an item, that item can be made accessible to subsequent transactions, rather than remaining locked until the update can be completed. A polyvalue describes the possible values that may be correct for an item, depending on the outcome of transactions that have been interrupted by failures. Frequently, the most important effects of a transaction (such as the payment of money) can be determined without knowing the exact values of the items in the data base. A polyvalue for an item that is accessed by such a transaction may be sufficient to determine such effects. By using polyvalues, we can guarantee that a data item will not be made inaccessible to any failure other than a failure of the site that holds the item. A strong motivation for the development of these protocols is the desire that the individual sites of a distributed information system fail independently, and that a site or a group of sites be able to continue local processing operations when a failure has isolated them from the rest of the sites. Many of the previous coordination mechanism have only considered the continued operation of the sites that remain with the system to be important. Another motivating factor for the development of these protocols is the idea that in many applications, the processing to be performed exhibits a high degre