79 Lightning Talk Dancing with Cores A Path For FineGrained Core Positioning within Kubernetes
summary
A CPU control plane is under development to mix shared and pinned cores and later add isolated cores, with CPU control plane exposed via cluster primitives. Benchmarks run on bare metal using Ansible for provisioning and batch scheduling against Google microservices and DeathStarBench. Pinning services and isolating sockets fixes noisy‑neighbor effects, surfaces a database bottleneck, and raises core utilization from 40% to 78%. Plans include virtualized hosts and measuring throughput and latency at larger scale.
AI-written summary · about
chapters (12)
- 0:00Fine-grained core positioning in Kubernetes
- 0:42Limitations of core managers
- 1:26Automating benchmarking with Ansible
- 2:49Selecting cloud-native workloads
- 3:51Benchmarking microservices with pod placement
- 5:18Optimizing core utilization
- 6:21HPC and AI workload needs
- 7:24Benchmarking gaps and future work
- 8:07Q&A: Virtualization and core placement
- 9:12Q&A: Kubernetes testing infrastructure
- 10:38Q&A: Multi-tenancy and network resources
- 11:41Q&A: Network sharing and jitter
related talks
06m⚡ Lightning Talk: Colocating Data Workloads and Web Services on Kubernetes to Improve Resource Utilization
34mUnlocking Performance: Topology-Aware CPU Scheduling W...
05mProject Lightning Talk: CoHDI: Dynamic Hardware Composition for AI‑Era Kubernetes…
11mLightning Talk: The Adaptive Edge: Balancing Cost Efficiency, Usability, and Performance Tradeoffs
08mLightning Talk: Orchestrating Kubernetes Clusters on HPC Infrastructure
23mThis is The Way: A Crash Course on the Intricacies of Managing CPUs...
21mDo Not Disturb Mode: Fine-Tuning Resources for Latency Sensitive Workloads
06mLightning Talk: Cloud Native Design Patterns