Responsive parallel computation: bridging competitive and cooperative threading

Abstract

Competitive and cooperative threading are widely used abstractions in computing. In competitive threading, threads are scheduled preemptively with the goal of minimizing response time, usually of interactive applications. In cooperative threading, threads are scheduled non-preemptively with the goal of maximizing throughput or minimizing the completion time, usually in compute-intensive applications, e.g. scientific computing, machine learning and AI. Although both of these forms of threading rely on the same abstraction of a thread, they have, to date, remained largely separate forms of computing. Motivated by the recent increase in the mainstream use of multicore computers, we propose a threading model that aims to unify competitive and cooperative threading. To this end, we extend the classic graph-based cost model for cooperative threading to allow for competitive threading, and describe how such a cost model may be used in a programming language by presenting a language and a corresponding cost semantics. Finally, we show that the cost model and the semantics are realizable by presenting an operational semantics for the language that specifies the behavior of an implementation, as well as an implementation and a small empirical evaluation. CCS Concepts • Software and its engineering → Parallel programming languages; Concurrent programming languages; Functional languages

Fairness in responsive parallelism

“…Probably the most famous such result is Brent's Theorem [Brent 1974] which gives a schedule that is 2-optimal for standard DAGs without priorities or responsiveness requirements. Muller et al [2017Muller et al [ , 2018 proposed the prompt scheduling principle, which extends classic ideas like Brent's to prioritized DAGs. A prompt schedule is one that does not execute a thread if there is a higher-priority thread that is ready and not already being executed.…”

“…This scheduler is quite effective at balancing purely computational tasks, and so makes a good baseline for comparing the scaling of computation times. The second comparison is with the responsive scheduler of [Muller et al 2017], which was also based on Spoonhower's framework and schedules interactive workloads to optimize for both computation and response time, but only for programs with two priorities. For the comparison to Spoonhower Work Stealing (SWS), we use benchmarks consisting of only a compute kernel (because SWS was not designed to handle interactive programs).…”

“…6.3.2 Photo Viewer. Our photo viewer application, derived from a benchmark from prior work [Muller et al 2017], displays JPEG images from a local folder, and allows the user to scroll forward and backward, and to jump to images out of order. When a new image is selected, the next 8 images are decoded by low-priority threads.…”

“…For example, a game might perform compute-intensive tasks to implement visual effects while it also interacts with the user. Recent work by Muller et al [2017Muller et al [ , 2018 has developed programminglanguage theory for such interactive parallel applications by considering languages that combine interaction and parallelism, and by allowing parallel tasks to have different priorities. Their work allows, for example, a parallel compute-intensive task to run in the łbackgroundž concurrently with a high-priority interactive task.…”

“…The prior work on responsive parallelism [Muller et al 2017[Muller et al , 2018 shows that it is theoretically possible to achieve both responsiveness and throughput. In this paper, we build on that work to account for fairness while maintaining provable guarantees on response time by presenting fairly prompt scheduling.…”