<!doctype html PUBLIC "-//W3C//DTD HTML 3.2 Final//EN"> <html> <head> <meta http-equiv="content-type" content="text/html; charset=iso-8859-1"> <meta http-equiv="content-style-type" content="text/css"> <link rel="stylesheet" type="text/css" href="style.css"> <title>ProGuard Results</title> </head> <body> <h2>Results</h2> <b>ProGuard</b> successfully processes any Java bytecode, ranging from small midlets to entire run-time libraries. It primarily reduces the size of the processed code, with some potential increase in efficiency as an added bonus. The improvements obviously depend on the original code. The table below presents some typical results: <p> <table> <tr> <th width="28%">Input Program</th> <th width="12%">Original size</th> <th width="12%">After shrinking</th> <th width="12%">After optim.</th> <th width="12%">After obfusc.</th> <th width="12%">Total reduction</th> <th width="12%">Time</th> <th width="12%">Memory usage</th> </tr> <tr> <td><a target="other" href="http://java.sun.com/j2me/">Worm</a>, a sample midlet from Sun's JME</td> <td align="center">10.3 K</td> <td align="center">9.8 K</td> <td align="center">9.6 K</td> <td align="center">8.5 K</td> <td align="center">18 %</td> <td align="center">2 s</td> <td align="center">19 M</td> </tr> <tr> <td><a target="other" href="http://www.javadocking.com/">Javadocking</a>, a docking library</td> <td align="center">290 K</td> <td align="center">281 K</td> <td align="center">270 K</td> <td align="center">201 K</td> <td align="center">30 %</td> <td align="center">12 s</td> <td align="center">32 M</td> </tr> <tr> <td><b>ProGuard</b> itself</td> <td align="center">648 K</td> <td align="center">579 K</td> <td align="center">557 K</td> <td align="center">348 K</td> <td align="center">46 %</td> <td align="center">28 s</td> <td align="center">66 M</td> </tr> <tr> <td><a target="other" href="http://www.clarkware.com/software/JDepend.html">JDepend</a>, a Java quality metrics tool</td> <td align="center">57 K</td> <td align="center">36 K</td> <td align="center">33 K</td> <td align="center">28 K</td> <td align="center">51 %</td> <td align="center">6 s</td> <td align="center">24 M</td> </tr> <tr> <td><a target="other" href="http://java.sun.com/javase/6/">the run-time classes</a> from Sun's Java 6</td> <td align="center">53 M</td> <td align="center">23 M</td> <td align="center">22 M</td> <td align="center">18 M</td> <td align="center">66 %</td> <td align="center">16 min</td> <td align="center">270 M</td> </tr> <tr> <td><a target="other" href="http://jakarta.apache.org/tomcat/index.html">Tomcat</a>, the Apache servlet container</td> <td align="center">1.1 M</td> <td align="center">466 K</td> <td align="center">426 K</td> <td align="center">295 K</td> <td align="center">74 %</td> <td align="center">17 s</td> <td align="center">44 M</td> </tr> <tr> <td><a target="other" href="http://www.kclee.com/clemens/java/javancss/">JavaNCSS</a>, a Java source metrics tool</td> <td align="center">632 K</td> <td align="center">242 K</td> <td align="center">212 K</td> <td align="center">152 K</td> <td align="center">75 %</td> <td align="center">20 s</td> <td align="center">36 M</td> </tr> <tr> <td><a target="other" href="http://ant.apache.org/">Ant</a>, the Apache build tool</td> <td align="center">2.4 M</td> <td align="center">401 K</td> <td align="center">325 K</td> <td align="center">242 K</td> <td align="center">90 %</td> <td align="center">23 s</td> <td align="center">61 M</td> </tr> </table> <p> Results were measured with ProGuard 4.0 on a 2.6 GHz Pentium 4 with 512 MB of memory, using Sun JDK 1.5.0 in Fedora Core 3 Linux. <p> The program sizes include companion libraries. The shrinking step produces the best results for programs that use only small parts of their libraries. The obfuscation step can significantly shrink large programs even further, since the identifiers of their many internal references can be replaced by short identifiers. <p> The Java 6 run-time classes are the most complex example. The classes perform a lot of introspection, interacting with the native code of the virtual machine. The 1500+ lines of configuration were largely composed by automated analysis, complemented by a great deal of trial and error. The configuration is probably not complete, but the resulting library successfully serves as a run-time environment for running applications like ProGuard and the ProGuard GUI. <p> For small inputs, timings are governed by the reading and parsing of the jars. For large inputs, the optimization step becomes more important. For instance, processing the Java 6 run-time classes without optimization only takes 2 minutes. <p> Memory usage (the amount of physical memory used by ProGuard while processing) is governed by the basic java virtual machine and by the total size of the library jars and program jars. <hr> <address> Copyright © 2002-2009 <a href="http://www.graphics.cornell.edu/~eric/">Eric Lafortune</a>. </address> </body> </html>