The Impact of Partial Orders among Strict and Probabilistic Quorums on the Data Consistency vs. Operation Availabilities Trade-Off

Abstract

Operation availabilities of pessimistic quorumbased replication strategies are strictly upper-bounded. Probabilistically relaxing their strict consistency notion permits to overcome this bound by introducing probabilistic data replication strategies. These strategies allow the exploitation of the data consistency vs. operation availabilities trade-off. In our approach, we add a certain number of only probabilistically intersecting quorums to the originally strictly intersecting partially ordered quorum system in order to "closely conserve" their original characteristic behavior. As one expects, the number and nature of probabilistic quorums added have a strong impact on the particular trade-off. But furthermore, it turns out that the order in which probabilistic and strict quorums must be probed is also very important. We relate two general strategies of ordering the probabilistic and strict quorums of those probabilistic quorum systems and evaluate their impact on the data consistency vs. operation availabilities trade-off. The evaluation uses Markov chain-based steady state analysis applied to representative instances of probabilistic quorum systems with the adequate partial orderings.

Relaxing data consistency along different dimensions for increasing operation availabilities

“…In contrast, in [11] the integration of the probabilistic quorums with the strict quorums is uniformly at random.…”

“…While these approaches address other quality measures or consider only one type of operation, the approach presented here builds on, completes and gives a 'complete picture' of the approaches of [10,11]. It is targeted towards increasing the read as well as write operation availability by relaxing consistency along different dimensions and analyses the role of these dimensions in the construction of PQSs.…”

“…The resulting poPQS first used the original strict quorums and only in case of their unavailability, the newly added probabilistic quorums were probed in their replication strategy-specific ordering. In contrast, in [11] we presented and analysed the partial ordering of the resulting poPQS which probes strict and probabilistic quorums in a uniformly random manner after the consideration of the replication strategy-specific ordering.…”

“…In [10,11], we introduced three concepts, (1) PQS generation based on an input poSQS, (2) probabilistic quorum selection driven by data consistency and (3) PQS integration of the resulting poPQS into the original poSQS. However, we analysed the two generation methods with the selection only in combination with the consistency-driven integration strategy first and then with the integration strategy with uniformly random choice between strict and probabilistic operations.…”

“…We analyse the data consistency versus operation availabilities trade-off for the two quorum generation methods and the selection introduced in [10]. But here, in contrast to [10,11], we additionally propose and analyse a third integration strategy that imposes partial ordering among strict and probabilistic quorums only within the priority classes derived from the replication strategy-specific partial ordering. Furthermore, we explain the quorum generation methods, the quorum selection and then continue with the probabilistic and strict quorum integration strategies, their comparison and their impact on the trade-off.…”