Re:Show HN: Kubestone – Benchmarking Operator for Kubernetes
by jacques_chester on Sunday November 17, 2019 @11:11AM
(#21558510)
How do you disentangle the performance effects of load generation from the target system? It looks at first glance to all be colocated in the same cluster.
[ Parent
| on HN ]
Re:Show HN: Kubestone – Benchmarking Operator for Kubernetes
by ottovonbivouac on Sunday November 17, 2019 @04:05PM
(#21560046)
You can use taints, tolerations, affinity and anti-affinity of k8s to schedule the load-generator to a different node than the one contains the service under test.
The load-gen and the service does not need to be co-located in case of REST (drill) and PostgreSQL (pgbench) based load testing.
In case of pure network benchmarking (with iperf & qperf) currently only the same cluster is supported.