Differentiated nonblocking: a new progress condition and a matching queue algorithm

TL;DR

This paper introduces a new progress condition called differentiated nonblocking, proposes a shared queue algorithm satisfying it, and demonstrates through implementation and experiments that it offers stronger guarantees and improved fairness with minimal throughput loss.

Contribution

The paper presents a novel progress condition and a matching queue algorithm, providing stronger guarantees and fairness improvements over existing solutions.

Findings

Stronger worst-case progress guarantees.

Significantly increased fairness.

Minimal throughput decrease.

Abstract

In this paper, we first propose a new liveness requirement for shared objects and data structures, we then give a shared queue algorithm that satisfies this requirement and we prove its correctness. We also implement this algorithm and compare it to a well-known shared queue algorithm that is used in practice. In addition to having a stronger worst-case progress guarantee, our experimental results suggest that, at the cost of a marginal decrease in throughput, our algorithm is significantly fairer, by a natural definition of fairness that we introduce here.