We performed a load test on both AWS NLB and ALB. On NLB, we reached up to 6.5KTPS throughput without any backend or response time issues. The response time of the API we are testing is less than 100ms.
We switched NLB to ALB and performed the same scenario. The throughput is not going beyond 2.5K TPS and it is very inconsistent unlike throughput in NLB.
We contacted the AWS support team, they suggested pre-warm the ALB before starting the load test. We prewarmed to support up to 10K tps.
We performed a load test after ALB pre-warm-up. We are facing the same throughput issue.
The connect time graph metrics of JMeter is showing that connect time is increasing linearly with load and it is taking up to 2-4 secs.
Kindly let me know what could be a potential bottleneck. It’s a simple standalone API, no database requests are performed.