2 ms·
In our production experience, utilization had far more to do with the service owners (or autoscalers/auto-tuners) correctly choosing the cgroups and CPU schedul
by dharmab 6y ago
In our production experience, utilization had far more to do with the service owners (or autoscalers/auto-tuners) correctly choosing the cgroups and CPU scheduler allocations, as well as the kernel settings for cgroup slicing and CPU scheduler. We had Mesos clusters with 3% utilization and have Kubernetes clusters with 95%+ utilization. But we also have Kubernetes clusters with <10% utilization.