Between Linearizability and Quiescent Consistency: Quantitative Quiescent Consistency

TL;DR

This paper introduces quantitative quiescent consistency (QQC), a relaxed correctness criterion for concurrent data structures that balances performance and correctness by adjusting relaxation based on contention levels.

Contribution

The paper defines QQC as a new relaxation of linearizability that scales with contention, providing a more practical correctness criterion for concurrent structures.

Findings

High-performance counters and stacks satisfy QQC

QQC refines quiescent consistency by allowing relaxation during contention

Three characterizations of QQC demonstrate its robustness and compositionality

Abstract

Linearizability is the de facto correctness criterion for concurrent data structures. Unfortunately, linearizability imposes a performance penalty which scales linearly in the number of contending threads. Quiescent consistency is an alternative criterion which guarantees that a concurrent data structure behaves correctly when accessed sequentially. Yet quiescent consistency says very little about executions that have any contention. We define quantitative quiescent consistency (QQC), a relaxation of linearizability where the degree of relaxation is proportional to the degree of contention. When quiescent, no relaxation is allowed, and therefore QQC refines quiescent consistency, unlike other proposed relaxations of linearizability. We show that high performance counters and stacks designed to satisfy quiescent consistency continue to satisfy QQC. The precise assumptions under which QQC holds provides fresh insight on these structures. To demonstrate the robustness of QQC, we provide three natural characterizations and prove compositionality.