Towards Container layer aware Scheduling Policies
summary
Implements two Kubernetes schedulers: ContainerLayer-Aware that picks nodes by shared container layers and a Multi-Objective policy that jointly reduces makespan, I/O and layer transfers. Uses a linear-program-based FOA scheduler run over simulated workloads built from FunctionBench and OpenWhisk traces on Grid5000, comparing against Kubernetes ImageLocality. Results show up to two orders of magnitude less layer data transferred, median energy cut ~50%, fewer machines used, and FOA runtime improved from minutes to seconds in the newer implementation.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 5:20Container layer sharing
- 13:53Experimental results
- 18:53Performance trade-offs
- 21:42Energy-aware scheduling
- 26:23Summary
- 28:11Q&A
related talks
47mRunning Multiple Schedulers in Kubernetes
39mPractical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at Netflix
30mSched.net: A Network-Aware Kubernetes Scheduler [I]
35mLeveraging the Kubernetes Enqueue Scheduling Plugin for Smarter Workload Placement
24mA K8s-Based Workload Allocation Optimizer for Minimizing Power Consumption
31mNew Cloud Foundry Architectures and Features and their Performance Impact
31mSIG Scheduling Update: Transition From Pod To Workload Scheduling
34mSIG-Scheduling Intro and Deep Dive