anvena / spring-vs-ejb-vs-cdi-benchmark

The JMH-based benchmark of Spring MVC vs EJB vs CDI RESTful web-service implementations

The simple JMH-based benchmark of Spring MVC vs EJB vs CDI RESTful web-service implementations

A very popular delusion in the Java world is the Spring Framework provides better performance than good old Enterprise Java Beans. The best approach to compare technologies is just to get a benchmark. This repository contains three implementations of the same simple RESTful web-service based on Spring MVC, EJB and CDI, respectively. The service design is the following: resource providers are injected into a "service" class and the "service" is injected into a controller. This is the usual design of an enterprise application: data-access objects are injected into services and the services are injected into facades or controllers. Different technologies are used for Dependency Injection as well as RESTfulization.

Let's have a look at the EJB version.

Resource:

@Stateless
public class EJBResourceA {

    public String message() {
        return "A#" + System.currentTimeMillis();
    }
}

Service:

@Stateless
public class MessageService {

    @EJB
    private EJBResourceA aresource;
    
    @EJB
    private EJBResourceB bresource;
    
    public String message() {
        return aresource.message() + bresource.message();
    }
}

REST Controller:

@Stateless
@Path("/")
public class MessageController {

    @EJB
    private MessageService service;
    
    @GET
    @Path("/message")
    @Produces({"text/plain"})
    public String message() {
        return service.message();
    }
}

The service just returns a string built by some letters and current time concatination:

JMH is a Java harness for building, running, and analysing nano/micro/milli/macro benchmarks written in Java and other languages targetting the JVM.

The benchmark code contained by the file.

Now it is time to destroy the myth!

---

Benchmark running

# java -jar rest-benchmark/target/rest-benchmarks.jar ".*Benchmark" -f 4 -wi 20 -i 20 -t 4 
-si true -gc true

Where:

• -f [int] - How many times to forks a single benchmark.
• -wi <int> - Number of warmup iterations to do.
• -i <int> - Number of measurement iterations to do.
• -si [bool] - Synchronize iterations?
• -t <int> - Number of worker threads to run with.
• -gc [bool] - Should JMH force GC between iterations?

Allowed parameters:

• server - application server's host, localhost is the default value.
• port - application server's port.
• path - path for the resource. Every path is relative to the /api/ catalog.
• implementation - technology used for the RESTful service implementation. The passed values are just substituted in the rest-ws-{implementation} template for web-applications context-roots.

Example:

# java -jar rest-benchmark/target/rest-benchmarks.jar ".*Benchmark" -f 4 -wi 20 -i 20 -t 4 
-si true -gc true -p port=14633 -p implementation=ejb,cdi

---

Results

Summary Diagram

The test results data - troughput - are normalized (the Spring Framework is 100%, bigger value is better):

EJB is up to 15% faster than the Spring Framework while CDI is up to 19% slower.

WebSphere Application Server 8.5.5.4 for z/OS

IBM Mainframe zBC 12: 2 General CP + 1 zIIP, 32 GB RAM, z/OS 2.1, IBM J9 VM R26 SR7 based on Oracle 7u55-b13

Operations per second, the picture is clickable:

WebSphere Application Server 8.5.5.8 Network Deployment

IBM Mainframe zBC 12: 2 IFL, 16 GB RAM, SUSE Linux Enterprise Server 12 s390x, IBM J9 VM R27 SR3 based on Oracle 7u85-b15

Operations per second, the picture is clickable:

WebSphere Application Server 8.5.5.8 Network Deployment

Lenovo ThinkPad T440: 2 Intel i5-4300U @ 1.90 GHz, 8 GB RAM, Windows 7 x64, IBM J9 VM R27 SR3 based on Oracle 7u85-b15

Operations per second, the picture is clickable:

Which result have you gone? Please, share it with me: [e-mail]