Java Magazine - November/December 2011

public List<Prediction> allPredictions(){

System.out.println("-- returning predictions"); return this.em.createNamedQuery( Prediction.findAll).getResultList();

}

See all listings as text

JAVA IN ACTION

Figure 4

average values. All the lines should be, on average, flat.

An increasing number of loaded classes might indicate problems with class loading and can lead to an OutOfMemoryError due to a shortage of PermGen space. An increasing number of threads indicates slow, asynchronous methods. A ThreadPool configured with an unbounded number of threads will also lead to an OutOfMemoryError. And a steady increase in memory consumption can eventually lead to an OutOfMemoryError caused by memory leaks.

VisualVM comes with an interesting profiling tool called Sampler. You can attach and detach to a running Java process with a little overhead and measure the most-expensive invocations or the size of objects (see Figure 4).

The sampling overhead is about 20 percent, so with an active sampler, you can still achieve 1,400 transactions per second. As expected, the application spends the largest amount of time communicating with the database.

measure the overhead, every invocation of the method allPredictions is logged with a System.out.println invocation, as shown in Listing 1.

Instead of 1,700 transactions per second, we are able to perform only about 800 transactions per second, as shown in Figure 5.

Let’s take a look at the VisualVM Sampler output shown in Figure 6. More time is spent in ThreadPrintStream .println() than in the most expensive database operation.

The actual EJB 3. 1 overhead is negligible. The $Proxy285.allPredictions() invocation is in the very last position and orders of magnitude faster than a single System.out.println.

Having a reference measurement makes identification of potential bottlenecks easy. You should perform stress tests as often as possible and compare the results. Performing nightly stress tests from the very first iteration is desirable. You will get fresh results each morning so you can start fixing potential bottlenecks.