Fairness in responsive parallelism

Muller, Stefan K.; Westrick, Sam; Acar, Umut A.

doi:10.1145/3341685

Cited by 5 publications

(6 citation statements)

References 68 publications

(54 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In contrast, Muller et al [24] design and prove responsiveness for a probabilistic scheduler of cooperative threads with priorities. The work is based on the cooperative suspension of threads but they interrupt threads that do not suspend; this prevents assumptions on the finiteness of local computation but compromises the guarantees of cooperative scheduling.…”

Section: :24mentioning

confidence: 98%

See 1 more Smart Citation

Provably Fair Cooperative Scheduling

Hähnle,

Henrio

2023

Programming

View full text Add to dashboard Cite

The context of this work is cooperative scheduling, a concurrency paradigm, where task execution is not arbitrarily preempted. Instead, language constructs exist that let a task voluntarily yield the right to execute to another task.The inquiry is the design of provably fair schedulers and suitable notions of fairness for cooperative scheduling languages. To the best of our knowledge, this problem has not been addressed so far.Our approach is to study fairness independently from syntactic constructs or environments, purely from the point of view of the semantics of programming languages, i.e., we consider fairness criteria using the formal definition of a program execution. We develop our concepts for classic structural operational semantics (SOS) as well as for the recent locally abstract, globally concrete (LAGC) semantics. The latter is a highly modular approach to semantics ensuring the separation of concerns between local statement evaluation and scheduling decisions.The new knowledge contributed by our work is threefold: first, we show that a new fairness notion, called quiescent fairness, is needed to characterize fairness adequately in the context of cooperative scheduling; second, we define a provably fair scheduler for cooperative scheduling languages; third, a qualitative comparison between the SOS and LAGC versions yields that the latter, while taking higher initial effort, is more amenable to proving fairness and scales better under language extensions than SOS.The grounding of our work is a detailed formal proof of quiescent fairness for the scheduler defined in LAGC semantics.The importance of our work is that it provides a formal foundation for the implementation of fair schedulers for cooperative scheduling languages, an increasingly popular paradigm (for example: akka/Scala, JavaScript, async Rust). Being based solely on semantics, our ideas are widely applicable. Further, our work makes clear that the standard notion of fairness in concurrent languages needs to be adapted for cooperative scheduling and, more generally, for any language that combines atomic execution sequences with some form of preemption.

show abstract

Section: :24mentioning

confidence: 98%

“…In rules (19) and (20) we must reset the counter to c ′ = 0. Finally, for (21), (24), and (23) the value of the counter does not matter, we arbitrarily choose c ′ = c.…”

Section: :32mentioning

confidence: 99%

Provably Fair Cooperative Scheduling

Hähnle,

Henrio

2023

Programming

View full text Add to dashboard Cite

show abstract

“…This research have shown that this bound is not purely theoretical and can in fact be realized efficiently in practice [1-3, 7, 12, 14, 16, 21, 23, 35], usually by using a variant of work stealing. Recent work has also made some progress extending this scheduling theory to account for latency [31] and competition among tasks, by allowing tasks to declare priorities [34,43].…”

Section: Related Workmentioning

confidence: 99%

DePa: Simple, Provably Efficient, and Practical Order Maintenance for Task Parallelism

Westrick¹,

Wang²,

Acar³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

A number of problems in parallel computing require reasoning about the dependency structure in parallel programs. For example, dynamic race detection relies on efficient "on-the-fly" determination of dependencies between sequential and parallel tasks (e.g. to quickly determine whether or not two memory accesses occur in parallel). Several solutions to this "parallel order maintenance" problem has been proposed, but they all have several drawbacks, including lack of provable bounds, high asymptotic or practical overheads, and poor support for parallel execution.In this paper, we present a solution to the parallel order maintenance problem that is provably efficient, fully parallel, and practical. Our algorithm-called DePa-represents a computation as a graph and encodes vertices in the graph with two components: a dag-depth and a fork-path. In this encoding, each query requires 𝑂 (𝑓 /𝜔) work, where 𝑓 is the minimum dynamic nesting depth of the two vertices compared, and 𝜔 is the word-size. In the common case (where 𝑓 is small, e.g., less than 100), each query requires only a single memory lookup and a small constant number of bitwise instructions. Furthermore, graph maintenance at forks and joins requires only constant work, resulting in no asymptotic impact on overall work and span. DePa is therefore work-efficient and fully parallel.

show abstract

“…Henry [28] gave a fair scheduler for processes on UNIX systems. Muller, Acar, and Harper [49] and Muller, Westrick, and Acar [50] studied fair scheduling for interactive computation and in the presence of priorities. Lahav et al [39] gave an account of process fairness under weak memory models.…”

Section: Fairnessmentioning

confidence: 99%

Fairness and Communication-Based Semantics for Session-Typed Languages

Kavanagh¹

2021

Preprint

View full text Add to dashboard Cite

Observed communication semantics [2] provide an intuitive notion of equivalence for communicating programs. We give the first observed communication semantics for a session-typed programming language with general recursion. This language, Polarized SILL [57,68], also supports channel and code transmission, synchronous and asynchronous communication, and functional programming. We present a framework inspired by testing preorders [16] for defining simulations based on observed communications. We show that the "external" simulations coincide with barbed precongruence.Polarized SILL is defined using a multiset-rewriting-style substructural operational semantics. To ensure that our observed communication semantics is well-defined, we introduce fairness for multiset rewriting systems. We construct a fair scheduler and we give sufficient conditions for traces to be fair.

show abstract

Fairness in responsive parallelism

Cited by 5 publications

References 68 publications

Provably Fair Cooperative Scheduling

Provably Fair Cooperative Scheduling

DePa: Simple, Provably Efficient, and Practical Order Maintenance for Task Parallelism

Fairness and Communication-Based Semantics for Session-Typed Languages

Contact Info

Product

Resources

About