How well does FastPak for Java Perform?

Recent benchmark tests on FastPak generally indicate comparable or superior performance when compared to using discrete JVMs to launch Java applications. The following table illustrates how two applications, FourierBenchmark and Analyzer, performed under FastPak and using dedicated JVMs for each application:


Processor Load Testing With A Single Instance of FourierBenchmark

(Lower results is better)


Operating System

Java Version

Number of Instances Running Simultaneously

FastPak Results (Time to complete 200 100K harmonic calculations)

Dedicated JVM Results (Time to complete 200 100K harmonic calculations)

Conclusions (which is faster, FastPak or Discrete JVM launches)

Windows XP

1.5

1

79.469 sec

81.641sec

FastPak by 2.7%

Windows XP

1.6

1

72.000 sec

77.516 sec

FastPak by 7.1%

OS X 10.4

1.5

1

57.221 sec

70.766 sec

FastPak by 19.1%

Suse Linux 10.1

1.5

1

91.833 sec

108.863 sec

FastPak by 15.6%

Suse Linux 10.1

1.6

1

83.473 sec

84.935 sec

FastPak by 1.7%



Processor Load Testing With 4 Simultaneous Instances of FourierBenchmark Running

(Lower results is better)


Operating System

Java Version

Number of Instances Running Simultaneously

FastPak Results (Average time to complete 200 100K harmonic calculations)

Dedicated JVM Results (Average time to complete 200 100K harmonic calculations)

Conclusions (which is faster, FastPak or Discrete JVM launches)

Windows XP

1.5

4

336.050 sec

304.542 sec

Discrete JVMs by 10.3%

Windows XP

1.6

4

284.996 sec

298.402 sec

FastPak by 4.5%

OS X 10.4

1.5

4

261.036 sec

279.424 sec

FastPak by 6.5%

Suse Linux 10.1

1.5

4

376.921 sec

414.941 sec

FastPak by 9.2%

Suse Linux 10.1

1.6

4

344.843 sec

339.522 sec

Discrete JVM by 1.6%



Graphics Load Testing With A Single Instance of Analyzer Running

(Higher result is better)


Operating System

Java Version

Number of Instances Running Simultaneously

FastPak Results (Number of iterations complete after 5 minutes)

Dedicated JVM Results (Number of iterations complete after 5 minutes)

Conclusions (which is faster, FastPak or Discrete JVM launches)

Windows XP

1.5

1

383

370

FastPak by 3.5%

Windows XP

1.6

1

388

382

FastPak by 1.6%

OS X 10.4

1.5

1

627

622

FastPak by 0.8%

Suse Linux 10.1

1.5

1

648

501

FastPak by 29.3%

Suse Linux 10.1

1.6

1

624

613

FastPak by 1.8%



Graphics Load Testing With 4 Simultaneous Instances of Analyzer Running

(Higher result is better)


Operating System

Java Version

Number of Instances Running Simultaneously

FastPak Results (Number of iterations complete after 5 minutes)

Dedicated JVM Results (Number of iterations complete after 5 minutes)

Conclusions (which is faster, FastPak or Discrete JVM launches)

Windows XP

1.5

4

147

156

Discrete JVMs by 5.8%

Windows XP

1.6

4

120

152

Discrete JVMs by 26.7%

OS X 10.4

1.5

4

152

144

FastPak by 5.6%

Suse Linux 10.1

1.5

4

183

138

FastPak by 32.6%

Suse Linux 10.1

1.6

4

164

143

FastPak by 14.7%



FourierBenchmark is an application extracted from one of FastPak's demo applications called FourierAnalysis. This extracted version has no GUI and calculates 100,000 harmonics for a digital pulse, repeating the set of calculations 200 times. When it starts, it records the time, and when complete it prints out the time in seconds it took to complete the operation. It makes extensive use of both mathematics and arrays. The source code for FourierBenchmark may be downloaded from the Downloads section of this web site as a jar file containing all the source code as well as a build script.

Analyzer is a demo application included with all distributions of FastPak. Analyzer is graphics intensive and it updates the screen with an array of data simulating a spectrum analyzer display. Each time it completes going through the array of analyzer data, a counter is incremented and it's displayed on the screen. It 's free running, meaning there are no timing delays embedded in the program, hence it's very graphics intensive and its speed will vary from system to system depending on the speed of both the overall system as well as the video card.

Please be aware that the test data above is not a comparison of operating systems or hardware platforms. The hardware configurations used in the tests are not realistically comparable to one another. The purpose of the tests was to illustrate that FastPak generally doesn't impair the performance of applications running underneath it, and can even improve performance of some applications even though it's consuming a fraction of the resources.