Although Java technology opens a wealth of opportunities for distributed scientific computing, the performance of Java needs to be evaluated to determine its suitability for numerical applications. A Java version of the Linpack Benchmark is available from Netlib The Linpack Benchmark is a numerically intensive test that has been used for years to measure the floating point performance of computers. The Java applet allows users to submit results for Java Linpack by filling in a form with information about the operating system and CPU of his local machine, as well as more detailed information about memory and processor speed, and whether or not the user's browser uses just-in-time (JIT) Java compilation. The user then presses a button of run the benchmark on his machine. The user's timings are then to the Java Linpack developers by email, who then update the timings and the graphical display that appear on the Java Linpack home page.
The test results are more a reflection of the state of Java systems than of the floating point performance of the underlying processors. Some Java systems do line by line interpretation and others perform ``just in time'' (JIT) compilation. As can be seen from the results, the JIT systems perform better, perhaps by an order of magnitude. The Linpack Java Benchmark allows the scientific user community to track the numerical performance of Java implementations over time over a range of architectures.