Probabilistically Bounded Staleness for Practical Partial Quorums

Abstract

Modern storage systems employing quorum replication are often configured to use partial, non-strict quorums. These systems wait only for a subset of their replicas to respond to a request before returning an answer, without guaranteeing that read and write replica sets intersect. While these partial quorum mechanisms provide only basic eventual consistency guarantees, with no limit to the recency of data returned, these configurations are frequently "good enough" for practitioners given their latency benefits. In this work, we discuss why partial quorums are often acceptable in practice by analyzing the staleness of data they return. Extending prior work on strongly consistent probabilistic quorums and using models of Dynamo-style anti-entropy processes, we introduce Probabilistically Bounded Staleness (PBS) consistency, which provides expected bounds on staleness with respect to both versions and wall clock time. We derive a closed-form solution for versioned staleness and model real-time staleness for representative Dynamo-style systems under internet-scale production workloads. We quantitatively demonstrate why, in practice, eventually consistent systems employing partial quorums often serve consistent data.

Benchmarking Eventual Consistency: Lessons Learned from Long-Term Experimental Studies

“…The difference in time between the last read of version n and the write timestamp of version n + 1 determines the client-centric t-Visibility [14] for systems with dirty reads (which is the normal case for cloud storage services and NoSQL systems).…”

“…All readers record detailed logs for each read containing tuples (readTimestamp;writeTimestamp;version). An offline analysis (or an online analysis lagging slightly behind) can then determine MRC violations in each individual reader's logs as well as the maximum inconsistency window, the probability of stale vs. fresh reads, and the version lag (k-Staleness [14]) based on all reader logs.…”

“…Bailis et al [33] use the distribution of network link latencies and quorum configurations to simulate consistency behavior for Dynamo-style [34] quorum systems. Their approach is, of course, limited to the usual bounds of simulation precision but offers a more cost-effective way for studying consistency behavior.…”

“…In the case of consistency guarantees and behavior, though, benchmarks and measurement approaches have only been developed over the last few years [7]- [13]. More recently, Bailis et al [14] presented an approach to also predict staleness behavior of certain classes of storage systems.…”

A Middleware Guaranteeing Client-Centric Consistency on Top of Eventually Consistent Datastores

“…This is certainly helpful in terms of information but has no influence on the actual guarantees of a datastore. The approach by Bailis et al [25] allows to give predictions for consistency guarantees of dynamo-style [3] quorum systems for scenarios where no failures occur. This again has no effect on the actual consistency guarantees of a storage system but at least increases transparency regarding QoS levels.…”