Pack-em or Stack-em: Advantages of Running OpenShift on Power
summary
OpenShift Container Platform (3.11) runs a MEAN-like workload (MongoDB Enterprise 4.2 + Node 14) on POWER9 and Skylake clusters with an SLA of <999 ms and a 1% error threshold. Tests use VM and bare-metal modes and JMeter load generators; POWER9 yields about 3.2 containers per core versus ~2.7 on Skylake in VMs, shows a ~2x price-performance advantage, and on bare metal scales to roughly 240 vs 100 containers, demonstrating higher container density and lower response times.
AI-written summary · about
chapters (18)
- 0:00Introduction and speaker intro
- 1:01IBM and Red Hat partnership
- 2:50Cloud workloads and vendor lock-in
- 4:38Application modernization with OpenShift
- 6:05Data-centric cloud applications
- 7:07Power processor family overview
- 8:54Workloads for OpenShift on Power
- 9:58Infrastructure modes for OpenShift
- 10:21Performance case study objective
- 14:37Software stack and architecture
- 16:23Data flow and cluster setup
- 19:14Performance comparison results
- 20:37Configuring compute units
- 21:40AI workloads and GPU acceleration
- 23:49Open hardware and software options
- 24:12Performance takeaways summary
- 25:15Live demo of OpenShift on Power
- 29:51Q&A session
related talks
31mPod scaling ad infinitum!!
20mBenchmarking HPC workloads on OpenShift
32mSoftware Defined Hardware Accelerated Data-Centric Computing...
10mLightning Talk: Does Cloud Elasticity Pay off for HPC Workloads?
06mPOWER your Cloud
28mOpenShift for the Enterprise or What Could Possibly Go Wrong
25mSustainability at the Core with Cloud Native Processors
30mOptimizing costs across several OpenShift clusters