node-vs-java-rest-api-performance

ReST API Performance Test: Java vs Node.js

Rumors are: Node.js is slow, because programming model is single thread, blabla

I tested the two!

My minimalistic (don't call it "micro") ReST Serice:

• HelloWorld GET service with one URL query-parameter
• similar simple POST API service with some not-too-complex marshaling involved

Tried to make the code as comparable as I can get:

• Java uses newest version of SpringBoot
• Node.js uses Express.js

...both on Alpine Docker, both w/o any hacks to make it faster

My Results ... you should proof by yourself

Also some kind of surprise for me!

Complexity

• Java has more files, more folders, more descriptors, hundreds LoC (= complexity)
• Node.js requires ONE file with 24 lines of code to do the same + package.json (14 lines)
• Java has slower round tips (code/compile/restart)
• Both run smoothly on Docker

Footprint

• Java + Springboot + Gradle is hungry: 900 MB RAM (for this tiny service!!) - I can reduce it to 400 MB w/o Gradle, buut ...
• Node.js is tiny: 6 MB at startup, 30 MB during load test

If you have lot micro-services on Kubernetes, this will cost you 30x more RAM! In my AKS clusters the scale is because of RAM (CPU is more than enough): So if you plan a HelloWorld cloud farm, Kubernetes cloud costs are x30 with Java!

Performance

Total (Avg):

• Node.js was able to process 2x more request per sec, than Java!! (each running alone on same server)

Details:

• Node.js serves 3x more GET requests than Java can do
• Java has faster response time
• Node.js and Java have similar requests/sec doing POST requests
• both Node.js and Java serves GET is 2x faster than POST: No surprise, GET-example it is a simpler request, transfers less data and has easier marshaling

Java and Node.js flood all CPU Cores with high load (against the rumors: N ode.js is single threaded and can't use multiprocessing, details see also here: https://nodejs.org/en/docs/guides/blocking-vs-non-blocking/)

Good news (both): 0% Errors

My Load Test Set Up:

REMARK: In JMeter you can to change the Port (8080 vs 8090, in both(!!) WebTests) to toggle between the two services.

• Servicess: 4 Core Server, Docker
• JMeter: separate 4 Core Server, running lot minutes (to eliminate ramp up etc.), even during test lot free RAM
  • JMeter was configured to run 100 Threads GET and POST mixed

Both running Ubuntu 18 LTS

To simulate bad networks, I recommend home WIFI, pls aks your kids to stream some videos :-D