DO Hackathon: Deployment Benchmark

Nobody likes waiting for a website to load, so it's important to have performance in mind while hosting websites.

In this article, I will compare the performance of both the deployment methods from the previous article in this series.

Benchmark Methodology

I'll be focusing exclusively on latency as that's what the end-user cares about. I'll be running an http benchmark tool from DigitalOcean droplets located in Toronto 1 and in Frankfurt 1, running these scripts:

# Kubernetes deployment
echo "GET https://krasnov.dev" | vegeta attack -duration=60s | vegeta report

# App Platform deployment
echo "GET https://denniskrasnov.com" | vegeta attack -duration=60s | vegeta report

The Results

Toronto 1 -> Kubernetes (Toronto 1)

Note that both resources are in the same region.

Requests      [total, rate, throughput]         3000, 50.02, 50.01
Duration      [total, attack, wait]             59.983s, 59.98s, 2.885ms
Latencies     [min, mean, 50, 90, 95, 99, max]  1.715ms, 3.944ms, 2.485ms, 3.147ms, 5.321ms, 50.502ms, 144.859ms
Bytes In      [total, mean]                     6348000, 2116.00
Bytes Out     [total, mean]                     0, 0.00
Success       [ratio]                           100.00%
Status Codes  [code:count]                      200:3000

Toronto 1 -> App Platform

Requests      [total, rate, throughput]         3000, 50.02, 50.01
Duration      [total, attack, wait]             59.986s, 59.98s, 5.764ms
Latencies     [min, mean, 50, 90, 95, 99, max]  5.141ms, 9.251ms, 7.443ms, 13.194ms, 17.124ms, 30.828ms, 223.472ms
Bytes In      [total, mean]                     40320000, 13440.00
Bytes Out     [total, mean]                     0, 0.00
Success       [ratio]                           100.00%
Status Codes  [code:count]                      200:3000

Frankfurt 1 -> Kubernetes (Toronto 1)

This represents a user that lives far away.

Requests      [total, rate, throughput]         3000, 50.02, 49.94
Duration      [total, attack, wait]             1m0s, 59.98s, 96.21ms
Latencies     [min, mean, 50, 90, 95, 99, max]  95.502ms, 103.55ms, 96.419ms, 98.411ms, 103.918ms, 344.845ms, 616.864ms
Bytes In      [total, mean]                     6348000, 2116.00
Bytes Out     [total, mean]                     0, 0.00
Success       [ratio]                           100.00%
Status Codes  [code:count]                      200:3000

Frankfurt 1 -> App Platform

Requests      [total, rate, throughput]         3000, 50.02, 50.01
Duration      [total, attack, wait]             59.986s, 59.98s, 6.333ms
Latencies     [min, mean, 50, 90, 95, 99, max]  4.248ms, 9.539ms, 6.037ms, 9.089ms, 11.87ms, 62.803ms, 666.033ms
Bytes In      [total, mean]                     40320000, 13440.00
Bytes Out     [total, mean]                     0, 0.00
Success       [ratio]                           100.00%
Status Codes  [code:count]                      200:3000

Mean latencies from all runs

Kubernetes

Kubernetes clusters are designed to be hosted in one location. This works well locally, but is slower when users are far away.

App Platform

Hosting on App Platform is geographically distributed (thanks to the built in CDN), and is performant no matter where the users are located.

Conclusion

For static websites, hosting on Digital App Platform can be more performant for users far away.