aks-jvm-benchmark

This project is used as an exercise for evaluating different deployment styles (horizontal scaling versus vertical scaling) of JVM workloads on Kubernetes.

Generate HdrHistogram chart

See: http://hdrhistogram.github.io/HdrHistogram/plotFiles.html

Demo script

First we show difference between GCs and Heap config with the same resource limits (1 CPU, 1 GB RAM).

Then we use the Load Balancer and fire up wrk against them.

Comparing different JVM settings

Start the benchmark with this script:

wrk -t10 -c50 -d5m -R3000 -L http://internal-sampleapp-all.default.svc.cluster.local/json

Wait with Prime Factor:

wrk -t10 -c50 -d5m -R3000 -L  http://internal-sampleapp-all.default.svc.cluster.local/waitWithPrimeFactor?duration=50\&number=927398173993974

Comparing different resource configurations

Because the k8s Load Balancer only round robin against pods, we must use Nginx for a two-tier load balancing approach.

Delete the gc-related pods, and deploy the redistribution pods, followed by nginx pod.

Start the benchmark with this script:

wrk -t10 -c50 -d5m -R3000 -L http://internal-nginx.default.svc.cluster.local/json

Wait with Prime Factor:

wrk -t10 -c50 -d5m -R3000 -L  http://internal-nginx.default.svc.cluster.local/waitWithPrimeFactor?duration=50\&number=927398173993974