Realistic Kong 3.8.0.0 Performance Benchmark

Why Realistic Performance Results matters

Kong does publishes a performance benchmark in the documentation. Which is a fantastic result to know.

This benchmark does not work for many clients. Two of the top reasons why a realistic performance is worth knowing for clients.

• Kong benchmark was run on AWS using EC2, single dedicated instance of c5.4xlarge which has 16 vCPU and 32 GiB memory. Clients often allocate 2 vCPU and 4 GiB memory for instances on cloud providers. These are significantly different machines in terms of compute and performance.

• Client often allocate smaller machines per node. Clients often look at scaling out Kong Instances rather than scaling up Kong Instances. So being closer to 2 vCPU and 4 GiB memory is more realistic performance result.

Realistic Config

Listed is realistic Kong machine config used for this benchmark results

CPU	2 CPU of Intel(R) Xeon(R) CPU @ 2.20GHz
Memory	4 GiB
Cloud Provider	GCP
Machine Type	e2-medium VM
OS	Ubuntu 20.04

Low Latency and Highest Throughput Performance Target

The following targets is set and configured

• 2 threads to make requests to match the 2 CPUs

• 10 concurrent connections to run the test

• Maximum average latency allowed is 5 ms

• Ignoring the max Latency and Standard Deviations

What is the maximum throughput that can be extracted from the compute allocated with the performance targets defined above from Kong.

Kong Configuration

1 proxy service

1 route

No plugins

ulimit set to 4096

proxy_access_log=off

log_level=debug

Kong configuration was loaded using declarative_config i.e. in DBless mode

Network Setup

The load runner, Kong Instance and Backend were hosted in the same GCP region and with in the same network.

Application Config

Backend

A HTTPS endpoint hosting a REST API with a GET

Returning a simple json response with very few characters. Similar to

{"hello":"world"}

Kong

Kong hosts a HTTPS endpoint to service as a REST API endpoint

Kong to backend is configured with HTTPs protocol

Load Runner

Load runner calls Kong endpoint using HTTPS

Performance Metrics

Number of Requests processed per Minute	Number of Requests Processed per Second	Average Latency
455,956	7600	4.32ms

What does it Mean

This is a measure of what is the maximum that could be extracted from a realistic compute allocated that is closer to typical customer deployment.

This is a point of reference for customers to use.