spring

Tomcat vs. Jetty vs. Undertow: Comparison of Spring Boot Embedded Servlet Containers

With the rise in popularity of micro services we have seen a similar rise in popularity of applications with embedded servlet containers. Spring boot is a Java based framework that supports application services. It runs as a standalone jar with an embedded servlet container or as a WAR file inside a container.
 
 
 
 
 
 
 
 

 
In this example, we will focus on the standalone jar with embedded servlet containers. The framework supports three different types of embedded servlet containers: Tomcat (default), Jetty and Undertow. We will compare the three and look at differences in properties, settings, performance and memory. Keep in mind that this example is analyzing the default configuration. There are many ways to optimize the performance or memory usage including to customize the auto configuration and component scanning.

We used Eclipse Neon, Java 8, Maven 3.3.9, Spring 1.4.3, Tomcat 8.5.6, Jetty 9.3.14 and Undertow 1.3.24.

1. Setup Spring Boot Application

We will use Maven to setup a new project in Eclipse with the appropriate dependencies. We will use the starter parent for this example but the dependencies in a production application will likely be altered to streamline, optimize or customize.

1.1 Setup Spring Boot Dependencies

The default embedded servlet container is Tomcat. This version of Spring Web 1.4.3 brings in Tomcat version 8.5.6.

pom.xml

<parent>
   <groupId>org.springframework.boot</groupId>
   <artifactId>spring-boot-starter-parent</artifactId>
   <version>1.4.3.RELEASE</version>
</parent>

<dependencies>
   <!-- TOMCAT -->
   <dependency>
      <groupId>org.springframework.boot</groupId>
      <artifactId>spring-boot-starter-web</artifactId>
   </dependency>
</dependencies>

1.2 Setup Spring Boot Main Application and Controllers

To setup the Spring Boot application you include the @SpringBootApplication annotation in your Main class. The @SpringBootApplication annotation brings in @SpringBootConfiguration, @EnableAutoConfiguration and @ComponentScan annotations.

Application.java

@SpringBootApplication
@ConfigurationProperties
public class Application {
public static void main(String[] args) {
   SpringApplication.run(Application.class, args);
}

You may choose to eliminate this annotation and add the @SpringBootConfiguration alone or to another class that allows you to customize the configuration. The @ComponentScan will scan your application for items like the @Controller you will need to setup a RESTful service. The following controller will return a simple “Hello World” string from a HTTP GET request. We have also included in the bundled example another endpoint mapping that returns a complex object type.

SampleController.java

@Controller
public class SampleController {

@Autowired
private ResourceLoader resourceLoader;

@RequestMapping("/")
@ResponseBody
public String home() {
   return "Hello World!";
}

1.3 Key Configuration Parameters

The default properties for all the embedded servlet containers are the same. Some of the most important properties to consider are the properties for configuring startup information like ports and application name, TSL, access logs, compression and many more.

For example, to configure SSL add the following to key value pairs to the application.properties.

application.properties

server.port=8443
server.ssl.key-store=classpath:keystore.jks
server.ssl.key-store-password=secret
server.ssl.key-password=another-secret

1.4 How to Find Additional Parameters

To explore the parameters for Spring boot applications you can add the Spring actuator dependency and the @ConfigurationProperties annotation to your Main class. You then visit the /configprops endpoint on your application to get a list of the available properties.

Application.java

@SpringBootApplication
@ConfigurationProperties
public class Application {

pom.xml

<dependency>
   <groupId>org.springframework.boot</groupId>
   <artifactId>spring-boot-starter-actuator</artifactId>
</dependency>
http://localhost:8080/jcg/service/configprops

1.5 Change version of Embedded Servlet Containers

The embedded servlet container versions are defined in the following parent dependency from the pom. You can change the version of the embedded servlet container by explicitly including the dependency and identifying a new version in the pom. We will show you how in the examples below.

pom.xml

<dependency>
   <groupId>org.springframework.boot</groupId>
   <artifactId>spring-boot-dependencies</artifactId>
   <version>1.3.7.RELEASE</version>
</dependency>

2. Tomcat

As Tomcat is the default embedded servlet container there is nothing you need to do to the default implementation to use Tomcat. You can change the version of Tomcat you are using or change properties in the pom.xml or application.properties files.

2.2 Change Version of Tomcat

pom.xml

<properties><tomcat.version>8.5.6</tomcat.version></properties>

<dependency>
   <groupId>org.apache.tomcat.embed</groupId>
   <artifactId>tomcat-embed-core</artifactId>
   <version>${tomcat.version}</version>
</dependency>
<dependency>
   <groupId>org.apache.tomcat.embed</groupId>
   <artifactId>tomcat-embed-el</artifactId>
   <version>${tomcat.version}</version>
</dependency>
<dependency>
   <groupId>org.apache.tomcat.embed</groupId>
   <artifactId>tomcat-embed-websocket</artifactId>
   <version>${tomcat.version}</version>
</dependency>

3. Jetty

To change the embedded servlet container to Jetty you need to edit the pom file to remove the Tomcat dependency and add Jetty.

3.1 Change to Jetty (version 9.3.14)

pom.xml

<dependency>
   <groupId>org.springframework.boot</groupId>
   <artifactId>spring-boot-starter-web</artifactId>
   <exclusions>
      <exclusion>
         <groupId>org.springframework.boot</groupId>
         <artifactId>spring-boot-starter-tomcat</artifactId>
      </exclusion>
   </exclusions>
</dependency>
<dependency>
   <groupId>org.springframework.boot</groupId>
   <artifactId>spring-boot-starter-jetty</artifactId>
</dependency>

4. Undertow

To change the embedded servlet container to Undertow you need to edit the pom file to remove the Tomcat dependency and add Undertow.

4.1 Change to Undertow (version 1.3.24 final)

Notice the undertow version included in the spring boot starter is incorrect, referring to 1.3.25. You’ll need to change it to 1.3.24.Final for this to work at the time of this article.

pom.xml

<dependency>
   <groupId>org.springframework.boot</groupId>
   <artifactId>spring-boot-starter-web</artifactId>
   <exclusions>
      <exclusion>
         <groupId>org.springframework.boot</groupId>
         <artifactId>spring-boot-starter-tomcat</artifactId>
      </exclusion>
   </exclusions>
</dependency>
<dependency>
   <groupId>org.springframework.boot</groupId>
   <artifactId>spring-boot-starter-undertow</artifactId>
</dependency>
<dependency>
   <groupId>io.undertow</groupId>
   <artifactId>undertow-core</artifactId>
   <version>1.3.24.Final</version>
</dependency>
<dependency>
   <groupId>io.undertow</groupId>
   <artifactId>undertow-servlet</artifactId>
   <version>1.3.24.Final</version>
</dependency>

5. Performance and Load

In this example, we will analyze both the peformance of HTTP requests and the memory footprint at startup of all three embedded servlet containers. We used JMeter to measure performance by simulating load and JVisualVM to look at the memory footprint.

5.1 Measure Performance

In this example, we will analyze both the peformance of simple RESTFul GET requests that return a string and more complex GET requests that return complex JSON objects. JMeter is the tool used to measure the performance of the the three different types of containers. The key to setting up this test was establishing thread groups with the appropriate load, a counter to dynamically update the input to the API and report viewers to display or aggregate the results. For the simple string examples, we used a thread group with 1000 threads that would loop 3 times through the sequence. It also used a ramp up time of 10 seconds. For the complex object examples, we used the same parameters but did not loop.

JMeter Tomcat Thread Group

 

JMeter Tomcat Summary Report

5.1.1 Tomcat

5.1.1.1 Simple String
Label# SamplesAverageMinMaxStd. Dev.Error %ThroughputReceived KB/secSent KB/secAvg. Bytes
Startup30007154935.783743610293.858360355.9593557255.67238466195
Others300010451.3596616820287.880241854.8209444954.53981144195
Others300010241.1550322750292.112950355.6269778555.3417113195
5.1.1.2 Complex Object with Dynamic Data
Label# SamplesAverageMinMaxStd. Dev.Error %ThroughputReceived KB/secSent KB/secAvg. Bytes
Startup100011431601322.8671905097.68486861202.333599919.937634322121
Others100032171.328216473097.88566954202.749516719.97861812121
Others100021161.110529603098.52216749204.067887920.108528332121
Others100021211.344498419098.53187506204.087995120.110509662121

5.1.2 Jetty

5.1.2.1 Simple Object
Label# SamplesAverageMinMaxStd. Dev.Error %ThroughputReceived KB/secSent KB/secAvg. Bytes
Startup30007056140.137050650291.516859456.082833355.22878197
Others300010211.0589250310293.599530256.4835033855.6233485197
Others300010210.9260343170294.348508656.6275939555.7652448197
5.1.2.2 Complex Object with Dynamic Data
Label# SamplesAverageMinMaxStd. Dev.Error %ThroughputReceived KB/secSent KB/secAvg. Bytes
Startup100011031397278.7961107098.13542689203.362671719.933758592122
Others100032201.500210319098.48335631204.083673920.004431752122
Others100032452.729377218098.29942003203.702509119.967069692122

5.1.3 Undertow

5.1.3.1 Simple Object
Label# SamplesAverageMinMaxStd. Dev.Error %ThroughputReceived KB/secSent KB/secAvg. Bytes
Startup30006045131.61887020295.683027863.8144034656.01807363221
Others300010221.2554478620292.740046863.1792483955.46051669221
Others300010181.5594779750294.377391863.5326206955.77071681221
5.1.3.2 Complex Object with Dynamic Data
Label# SamplesAverageMinMaxStd. Dev.Error %ThroughputReceived KB/secSent KB/secAvg. Bytes
Startup10007031114197.1333241097.059109203.396936119.620442012145.893
Startup1000423852132.6443576098.02960494205.632413520.007995542148
Others100032191.293570253098.55129595206.630500420.018231992147
Others100022271.659250132098.74592673207.038578820.057766372147
Others100021171.260904041098.28975821206.082139519.965107142147

 

5.2 Measure Memory

To measure the memory of each embedded servlet container we looked at the memory usage on startup. JVisualVM is a tool provided with the Java Development Kit for visualizing the memory and footprint of java applications. We used this tool to show the initial startup impact of each of the three embedded servlet containers. The heap size and thread counts are key in analyzing this initial footprint. The ten threads that are common to all three containers include: JMX server connection timeout, RMI Scheduler, RMI TCP Connection (2), RMI TCP Accept, Attach Listener, DestroyJavaVM, Signal Dispatcher, Finalizer and Reference Handler.

JVisualVM Report

5.2.2 Tomcat

Heap Size: 697,827,328 B
Used: 124,260,976 B
Max: 2,147,483,648 B

Threads: 17 Live, 22 Started

5.2.3 Jetty

Heap Size: 628,621,312 B
Used: 311,476,776 B
Max: 2,147,483,648 B

Threads: 19 Live, 22 Started

5.2.4 Undertow

Heap Size: 630,718,464 B
Used: 114,599,536 B
Max: 2,147,483,648 B

Threads: 17 Live, 20 Started

6. Compare

6.1 Performance

While all three of the embedded servlet containers had similar performance under the parameters used in this example, Undertow seems to have the best performance with Tomcat and Jetty close behind. The memory footprint of Jetty on startup was the largest using 311 MB. Tomcat and Undertow had similarly low initial footprints around 120 MB with Undertow coming in the lowest at 114 MB. The key difference in the response headers is that Undertow includes HTTP Persistent connections by default. This header will be used in clients that support persistent connections to optimize performance by reusing connection details.

6.1.1 Tomcat Response Headers

Content-Type →application/json;charset=UTF-8
Date →Mon, 09 Jan 2017 02:23:26 GMT
Transfer-Encoding →chunked
X-Application-Context →JcgSpringBootContainers:# Application index.

6.1.2 Jetty Response Headers

Content-Type →application/json;charset=UTF-8
Date →Mon, 09 Jan 2017 02:29:21 GMT
Transfer-Encoding →chunked
X-Application-Context →JcgSpringBootContainers:# Application index.

6.1.3 Undertow Response Headers

Connection →keep-alive
Content-Type →application/json;charset=UTF-8
Date →Mon, 09 Jan 2017 02:20:25 GMT
Transfer-Encoding →chunked
X-Application-Context →JcgSpringBootContainers:# Application index.

7. Conclusion

The numbers indicate that Undertow is the best in performance and memory usage. It is encouraging to see that Undertow is embracing the latest capabilities and defaulting to persistent connections. The numbers do not indicate a dramatic difference in performance based on the load used in this example but I would imagine that they would scale and that if performance is the most important factor Undertow is the right match for your application. It is also reasonable to think that an organization may favor an embedded servlet container because of familiarity with it’s capabilities. Many times the defaults settings will have to change because of application requirements that include performance, memory usage and functionality.

8. Download the Source Code

Here we compared three types of embedded servlet containers you can include in a Spring Boot Application.

Download
You can download the Eclipse project here: JcgSpringBootContainers

Andy Beck

Andy is a Senior Software Engineer that has sixteen years of experience working on web, desktop and mobile applications. He holds a Bachelor and Master's degree in Computer Science from the University of Virginia and George Washington University respectively. He specializes in working with Java web services and has significant experience working web applications, databases and continuous integration and deployments. He is currently working as a technical lead at a financial technology organization where he supports mobile application services in Java.
Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Learn how your comment data is processed.

1 Comment
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Nicolò Scarpa
Nicolò Scarpa
3 years ago

Nice article.

Didn’t know of the existance of JVisualVM.

Thanks!

Back to top button