Most distributed systems proposed on the basis of the concept of atomic action or transaction strongly limit parallelism, thus reducing their level of efficiency. In this paper, features of efficiency in a distributed...
详细信息
Most distributed systems proposed on the basis of the concept of atomic action or transaction strongly limit parallelism, thus reducing their level of efficiency. In this paper, features of efficiency in a distributed transaction system are investigated.
Decentralized consensus protocols are described by successive rounds of message interchanges. protocols that achieve a consensus in one round of message interchange need messages of a given set of sites in the system....
详细信息
Decentralized consensus protocols are described by successive rounds of message interchanges. protocols that achieve a consensus in one round of message interchange need messages of a given set of sites in the system. Here, a communication scheme is presented, based on finite projective plans, which needs only a given set of messages for each round. Employing this communication scheme, decentralized consensus protocols are developed that achieve a consensus within 2 rounds of message interchange. The protocols are symmetric, and the communication plan does not impose any hierarchical structure. The scheme is illustrated using blocking and nonblocking commit protocols, decentralized extrema finding, and computation of the sum function.
We address the problem of maintaining the distributed database consistency in presence of failures while maximizing the database availability. Network Partitioning is a failure which partitions the distributed system ...
详细信息
We address the problem of maintaining the distributed database consistency in presence of failures while maximizing the database availability. Network Partitioning is a failure which partitions the distributed system into a number of parts, no part being able to communicate with any other. Formalizations of various notions in this context are developed and two measures for the performances of protocols in presence of a network partitioning are introduced. A general optimality theory is developed for two classes of protocols—centralized and decentralized. Optimal protocols are produced in all cases.
Designing atomic commit protocols for a distributed database, supported by a distributed database management system, is a difficult task. The goal is to find protocols that ensure atomicity in the case of inappropriat...
详细信息
Designing atomic commit protocols for a distributed database, supported by a distributed database management system, is a difficult task. The goal is to find protocols that ensure atomicity in the case of inappropriate and perhaps repetitive failures. A model is used to prove existence results about resilient protocols for site failures that do not partition the network and then for those that do. Independent recovery, a scheme where a recovering site makes a transition directly to a final state without communicating with other sites, is a pessimistic recovery technique for site failures. It uses local state information available at the time of failure. This strategy is only resilient to a single failure. The database is left in an inconsistent state if independent recovery is tried after failures happening at the same time. Therefore, a recovering site cannot determine when it is safe to use independent recovery. A partitioned network results from at least 2 sites that are unable to communicate with each other. In the pessimistic case where messages are lost, no nonblocking protocols exist. The best protocols allow no more than one group of sites to continue while the remaining groups block. For all cases, fundamental limitations on protocol resiliency are derived.
暂无评论