jsfbench

Inspired by https://github.com/lu4242/performance-comparison-java-web-frameworks

I took the JSF application, ported it to CDI and created profiles for MyFaces 4 and Mojarra 4. The JMeter config is preconfigured for 100 threads and 200 loops.

Its also quite interesting compared to another case (http://tandraschko.blogspot.com/2019/03/way-to-myfaces-30.html), where the results of Mojarra 2.3 and MyFaces 2.3 were almost the same and MyFaces 2.3-next was ~15% faster as both Mojarra 2.3 and MyFaces 2.3. The big difference is that this benchmark simulates a real-world application with Hibernate and a bit business logic.

NOTE: MyFaces and Mojarra are both configured for better performance. It does not test stateless view or MyFaces ViewPooling. There should not be a big difference.

Results 2024-03-09

	Average	Median	90th pct	Throughput
Mojarra 4	2.80ms	2.00ms	3.00ms	4373.13
MyFaces 4	1.18ms	1.00ms	2.00ms	4526.78

How to run it?

Currently only tested on Windows 11 + JDK 11 (Adoptium)

Configure Tomcat

1. download Tomcat 10.1.X
2. extract it 2 times (tomcat-myfaces, tomcat-mojarra)
3. copy the server.xml file from ./tomcat/conf to all 2 instances, which disables autoDeployment and accessLogValve

Build the application

1. go the ./application
2. mvn clean package -Pmyfaces, copy the generated war and extract it to ./tomcat-myfaces/webapps/jsfbench
3. mvn clean package -Pmojarra, copy the generated war and extract it to ./tomcat-mojarra/webapps/jsfbench

Run via JMeter

1. download JMeter 5.6+
2. Increase MaxUserPort if you are on Windows: https://deploymentresearch.com/research/post/532/fix-for-windows-10-exhausted-pool-of-tcp-ip-ports
3. run one of the tomcats (e.g. ./tomcat-myfaces/bin/startup.bat)
4. run jmeter (./jmeter/bin/jmeter.bat)
5. open the ./application.jmx via JMeter
6. run it once for a warmup
7. clean the results
8. run again and see the results under Test Plan/Thread Group/Aggregate Grapp