Scheduler Workloads

The Linux Plumbers 2017 scheduler workload track will focus on understanding various workloads and their impact on the scheduler. Our objective is to initiate a cross organizational and architectural discussion involving currently available(or in development) benchmarks and their effectiveness in evaluating the scheduler for these workloads.

A reliable and scalable benchmark that can be run without too much setup overhead will always help in identifying scheduler issues for different architectures. Some of the questions we hope can be answered, whether we have all the benchmarks we need or new ones need to be developed. In the long run, this will help in achieving higher scalability and debug performance related issues easily.

Key Attendees (tentative)

  • Paul E. McKenney (confirmed)
  • Steven Rostedt (Confirmed)
  • Peter Zijlstra
  • Srikar Dronamraju (confirmed)
  • Tim Chen (confirmed)
  • Stephen Hemminger (confirmed)
  • Andi Kleen
  • Mike Galbraith
  • Paul Turner
  • Atish Patra (Confirmed)
  • Dhaval Giani (confirmed)
  • Chris Mason (confirmed)
  • Josef Bacik (confirmed)
  • Juri Lelli (confirmed)
  • Tejun Heo (confirmed)
  • Morten Rasmussenn (confirmed)
  • Dietmar Eggemann (confirmed)
  • Patrick Bellasi (confirmed)
  • Viresh Kumar (confirmed)
  • Vincent Guittot (confirmed)
  • Mathieu Desnoyers
  • Daniel Bristot de Oliveira (confirmed)
  • Rafael J. Wysocki (Confirmed)

Problem space

The following is the list of items presently considered as candidate topics for the microconf. Still very much subject to change.

Some of the workloads that can be discussed:

  • Database workload (Real database workloads takes longer time to setup, run and debug.A simulation program of intended behavior in Open Source C program would be great)
  • Interactive workload (mainly for desktops)
  • Network throughput workload (Networking applications)
  • Mobile workloads (usually require Android; power consumption very important alongside latency)

Some of the interesting benchmarks that can be discussed are described here.

Here are some details on the benchmarks:

rt-app: Source

schbench: Details Source

Proposed Topics

Please add topic proposals here:

  • Use rt-app to run mobile-like (and other types of) synthetic workloads (Juri Lelli mobile workloads usually require Android; remove the dependency by simulating such workloads with rt-app
  • Power alongside performance (latency) (Juri Lelli on the importance to check for power/energy regressions alongside performance; how to check across different archs/platforms?
  • RT applications on 10s of CPUs (Steven Rostedt The IPI work load of RT push/pull scheduling logic can cause IPI storms with RT tasks running on 10s of CPUs. Discuss ways to deal with this.
  • Using bpf to model application behavior (Josef Bacik Talk about work I've done using BPF to model large applications for building test cases for scheduler issues we hit in production. Discuss how the tool can be improved and used to help keep regressions from happening.
  • Test and develop running opensource applications (Paul E. McKenney Scheduler benchmarks and tests that can allow the scheduler developers and maintainers to check their work without having to have one each proprietary application.
  • Tools to analyze scheduler behavior based on tracing and scheduler model (Mathieu Desnoyers Event tracing, post-processing tools and scheduler models: how can we improve instrumentation and improve the tools/models to better understand the scheduler behavior ?
  • Workloads for evaluation CPU PM modifications ( Rafael J. Wysocki Discussion on which workloads to use (or what is recommended etc) for the evaluation of CPU PM changes in the kernel.

Tentative schedule


Runner: Atish Patra, Dhaval Giani, Juri Lelli

2017/scheduler_workloads.txt · Last modified: 2017/08/08 16:45 by
Except where otherwise noted, content on this wiki is licensed under the following license:CC Attribution-Noncommercial-Share Alike 3.0 Unported
Recent changes RSS feed Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki