Relaxing data consistency along different dimensions for increasing operation availabilities

Abstract

Pessimistic quorum-based data replication strategies have strictly upper-bounded operation availabilities. The probabilistic relaxation of their strict consistency notion along different dimensions permits to overcome this upper-bound. The resulting probabilistic replication strategies allow the exploitation of the data consistency versus operation availabilities trade-off. In our approaches, we add a certain number of probabilistically intersecting quorums to the originally strictly intersecting partially ordered quorum system in order to 'closely conserve' its original characteristic behaviour. As one expects, the number and nature of probabilistic quorums added have a strong impact on the particular trade-off. We present two algorithms for the generation of probabilistic quorums based on input strict quorums. Furthermore, the order in which probabilistic and strict quorums are to be probed and used for executing operations is also very important. We relate three general strategies of ordering the probabilistic and strict quorums of those probabilistic quorum systems (PQSs), evaluate their impact on the data consistency versus operation availabilities trade-off and compare them in terms of this trade-off. The evaluation uses Markov chain-based steady-state analysis applied to a representative instance of PQSs with adequate partial orderings.