Multi-resource Real-Time Reader/Writer Locks for Multiprocessors

Abstract

A fine-grained locking protocol permits multiple locks to be held simultaneously. In the case of real-time multiprocessor systems, prior work on such protocols has considered only mutex constraints. This unacceptably limits concurrency in systems in which some resource accesses are read-only. To remedy this situation, a variant of a recentlyproposed fine-grained protocol called the real-time nested locking protocol (RNLP) is presented that enables concurrent reads. Like the original RNLP, this reader/writer version is a "pluggable" protocol that has different variants for different schedulers and analysis assumptions. Some of these variants are asymptotically optimal with respect to priority-inversion blocking. This paper is the first to address the algorithmic and analytical intricacies that arise when attempting to support read-only resource accesses under fine-grained locking in multiprocessor real-time systems.

Formal Verification of Real-Time Autonomous Robots: An Interdisciplinary Approach

“…This duality between multi-resource locks and nested locking allows the real-time nesting locking protocol (RNLP) family ( Ward and Anderson, 2012 ; Ward and Anderson, 2013 ; Ward and Anderson, 2014 ; Ward, 2016 ) to provide a solution in both use cases. Dynamic group locks (DGLs) ( Ward and Anderson, 2013 ; Ward, 2016 ) are the multi-resource lock variant of RNLP, and reader–writer RNLP (R/W RNLP) ( Ward and Anderson, 2014 ) provides a reader–writer extension to DGL. With contention-sensitive RNLP (C-RNLP) ( Jarrett et al, 2015 ), there is also an extension to RNLP that relaxes the strict FIFO ordering and tries to dynamically eliminate transitive blocking chains .…”

“…Conceptually, RNLP locks are always presented by using dedicated queues per resource in the literature ( Ward and Anderson, 2012 ; Ward and Anderson, 2013 ; Ward and Anderson, 2014 ; Ward, 2016 ). However, later work of the authors gives a hint to single-queue implementations of the non-reader–writer variants ( Jarrett et al, 2015 ).…”

“…State-of-the-art fine-grained multi-resource protocols (from outside the robotic community) are promising candidates, yet none satisfies all the above three requirements. In particular, DGL, the multi-resource version of the real-time nesting locking protocol (RNLP) family ( Ward and Anderson, 2012 ; Ward and Anderson, 2013 ; Ward and Anderson, 2014 ; Ward, 2016 ), suffers from efficiency drawbacks and does not support mixed read–write requests, whereas MRLock ( Zhang et al, 2013 ) shows degraded predictability in corner cases ( Section 6.3 ). The core of this proposition is to benefit from the advantages of DGL and MRLock in order to propose a new implementation of a predictable-and-efficient locking protocol that is suitable for real-time robots.…”

Supporting Nested Resources in MrsP

“…The first proposal for fine-grained analysis was proposed in [10]; forcing a strict order on locks and releases (locking operations are not allowed after a release has been performed on the nesting). An extension of this work is the Real-time Nested Locking Protocol (RNLP) [12,13] which limits the concurrency on nested resource accesses by means of a token mechanism and provides a set of request satisfaction mechanisms aiming for optimality under different system configurations.…”

A contention-sensitive fine-grained locking protocol for multiprocessor real-time systems

“…As noted in Sec. 1, the RNLP is actually a family of protocols: a given instantiation is obtained by determining how waiting is realized (spinning or suspending), the progress mechanism to employ (priority inheritance, priority boosting, or priority donation), and potentially other factors (such as whether "long" critical sections are to be distinguished from "short" ones [12], and whether reader/writer sharing constraints exist [13]). All such instantiations share a common structure, which consists of two components: a token lock and a request satisfaction mechanism (RSM).…”