A Simple Improvement of the Work-stealing Scheduling Algorithm

Vrba, Zeljko; Halvorsen, Paal; Griwodz, Carsten

doi:10.1109/cisis.2010.67

Cited by 6 publications

(5 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The implication of the very different behaviors observed on the different microarchitectures is a demand for scheduling that can adapt to varying conditions using instrumentation data collected at runtime. Our next step is therefore to design such a low-level scheduler in the context of our P2G processing framework [7] using our improvements for the work-stealing approach as a starting point [17].…”

Section: Discussionmentioning

confidence: 99%

“…In this work, we have identified several challenges with respect to low level scheduling. The de facto standard is a variant of work-stealing scheduling [2], for which we have earlier proposed modifications [17]. We see challenges that have not been addressed by this, and in our work of designing an efficient scheduler, we investigate in this paper how to structure parallel execution of multimedia filters to maximize the performance on several modern architectures.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Low-Level Scheduling Implications for Data-Intensive Cyclic Workloads on Modern Microarchitectures

Espeland

Olsen

Halvorsen

et al. 2012

2012 41st International Conference on Parallel Processing Workshops

Self Cite

View full text Add to dashboard Cite

Abstract-Processing data intensive multimedia workloads is challenging, and scheduling and resource management are vitally important for the best possible utilization of machine resources. In earlier work, we have used work-stealing, which is frequently used today, and proposed improvements. We found already then that no singular work-stealing variant is ideally suited for all workloads. Therefore, we investigate in more detail in this paper how workloads consisting of various multimedia filter sequences should be scheduled on a variety of modern processor architectures to maximize performance. Our results show that a lowlevel scheduler additionally cannot achieve optimal performance without taking the specific micro-architecture, the placement of dependent tasks and cache sizes into account. These details are not generally available for application developers and they differ between deployments. Our proposal is therefore to use performance monitoring and dynamic adaption for the cyclic workloads of our target multimedia scenario, where operations are repeated cyclically on a stream of data.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Low-Level Scheduling Implications for Data-Intensive Cyclic Workloads on Modern Microarchitectures

Espeland

Olsen

Halvorsen

et al. 2012

2012 41st International Conference on Parallel Processing Workshops

Self Cite

View full text Add to dashboard Cite

show abstract

“…front [ 2] front [ 1] M task objects Figure 2. Data structure organization of our framework status of all task objects.…”

Section: Front[ N]mentioning

confidence: 99%

“…And two of the three are ubiquitous in all pipeline applications: first, any load imbalances in the pipeline reduce its performance; second, the concurrency in a pipeline is tightly coupled with the set of chosen stages, so the using of more processors may require the pipeline structure be changed to rebalance it. We use work-stealing [1][2][3] technique to implement a load balancing scheduler which would address these two problems. The work-stealing is a widely used load balancing algorithm.…”

Section: Introductionmentioning

confidence: 99%

A Non-blocking Programming Framework for Pipeline Application on Multi-core Platform

et al. 2011

2011 IEEE Ninth International Symposium on Parallel and Distributed Processing With Applications

View full text Add to dashboard Cite

Many applications meet certain programming patterns like pipeline, fork-join, do-all etc. While tools such as OS threads and OpenMP allow programmers only to express task or data parallelism, special support for programming patterns is distinctly lacking. Intel threading building blocks (TBB) is developed to address this problem, but its scheduler is general and not optimized for any of its parallel algorithms which include pipeline specially. In this paper, we provide a non-blocking framework for pipeline application on multi-core platform. We target linear pipeline in which each filter has one entrance and one exit. We design a novel work-stealing scheduler optimized specially for pipeline application: first, priority based stealing, priority is calculated for each filter in pipeline so that a worker can find the optimal "victim" easily when it needs to steal; second, multiple tasks can be stolen at a time so that much stealing time is reduced. A nonblock queue is used to store intermediate result to reduce lock overhead and increase scalability. We apply our framework to four case studies, including text filter, twofish, ferret, dedup. And our framework reduces execution time of TBB by 72% in best case and 20% on average on an 8 core machine.

show abstract

“…They are caused by high contention over run-queues when stealing. We have therefore modified the work-stealing algorithm so that an unblocked KP is placed on the last CPU that executed it [41] (WS-LAST). We do not use the non-blocking queue described in [4]; instead we use an ordinary mutex per run-queue.…”

Section: Scheduling Policiesmentioning

confidence: 99%

The Nornir run-time system for parallel programs using Kahn process networks on multi-core machines—a flexible alternative to MapReduce

et al. 2010

Self Cite

View full text Add to dashboard Cite

Even though shared-memory concurrency is a paradigm frequently used for developing parallel applications on small-and middle-sized machines, experience has shown that it is hard to use. This is largely caused by synchronization primitives which are low-level, inherently non-deterministic, and, consequently, non-intuitive to use. In this paper, we present the Nornir run-time system. Nornir is comparable to well-known frameworks such as MapReduce and Dryad that are recognized for their efficiency and simplicity. Unlike these frameworks, Nornir also supports process structures containing branches and cycles. Nornir is based on the formalism of Kahn process networks, which is a shared-nothing, message-passing model of concurrency. We deem this model a simple and deterministic alternative to shared-memory concurrency. Experiments with real and synthetic benchmarks on up to 8 CPUs show that performance in most cases scales almost linearly with the number of CPUs, when not limited by data dependencies. We also show that the modeling flexibility allows Nornir to outperform its MapReduce counterparts using well-known benchmarks.

show abstract

A Simple Improvement of the Work-stealing Scheduling Algorithm

Cited by 6 publications

References 11 publications

Low-Level Scheduling Implications for Data-Intensive Cyclic Workloads on Modern Microarchitectures

Low-Level Scheduling Implications for Data-Intensive Cyclic Workloads on Modern Microarchitectures

A Non-blocking Programming Framework for Pipeline Application on Multi-core Platform

The Nornir run-time system for parallel programs using Kahn process networks on multi-core machines—a flexible alternative to MapReduce

Contact Info

Product

Resources

About