Performing large-scale network testing on Red Hat OpenShift: A 100 Gbps approach
Red Hat News, Thursday, January 18th, 2024
This article aims to benchmark the results for 100 Gbps network interface assets and develop tools for driving the load.
Recently, a customer faced a network test issue in which the anticipated bandwidth was observed at the node level, but there was a reduction of 25% to 30% at the pod level. The customer could resolve the issue by using HostNetwork for the pods because it bypasses OVN-K hops. However, this approach can have security implications, so that's not the ideal solution.
The analysis further revealed that the customer application exhibited low latency and high packets per second (PPS), especially when routing traffic directly through the host network interface. The customer's environment had bare metal worker nodes, each equipped with two 100GB NICs. I tested the network with iPerf3 and validated the network speeds between pods. No automated procedure to conduct these tests existed. The customer is actively seeking a script to validate the scenario.