Java Software C++ Development Management Performance

Benchmarking Java Against C/C++ for Interactive

Scientific Visualization

Sudhir Sangappa K. Palaniappan Richard Tollerton

AVS Express Technical Team Dept. Comp. Eng & Comp. Science Dept. Comp. Eng & Comp. Science

Advanced Visual Systems Inc University of Missouri - Columbia University of Missouri - Columbia

Waltham, MA 02451 USA Columbia, MO 65211 USA Columbia, MO 65211 USA

******@***.*** ******@****.****.********.*** *******@****.****.********.***

environments co-developed with NASA Goddard Space Flight

ABSTRACT Interactive scientific visualization

Center has been an important tool for understanding

applications require both high performance graphics and

correlations between geophysical datasets and coupled

numerical computation capabilities. C/C++ programs are able

processes in the earth-ocean-atmosphere-biosphere system.

to take direct advantage of specialized graphics rendering

Complex phenomena are studied by scientists using NASA

hardware (referred to as a Graphics Processing Unit or GPU) for

Earth Observing System satellite imagery, timevarying

display and special SIMD instructions within the CPU for

multidimensional volumes from numerical models, synoptic

computation. The former using for example the OpenGL library

observations, etc. The Loop module of the DISS, implemented

and latter using compiler extensions. Software developed i n

in C is used for interactive animation, roaming, zooming, and

Java offers portability, robustness and security. However, the

probing of high resolution time sequence datasets. JLoop is a

graphics rendering performance, memory efficiency, and

high performance platform independent implementation of

computational speed of Java for visualization of large

Loop in Java and QtLoop is a standalone version in C++.

scientific data sets has not been well characterized.

JLoop is used to benchmark portability, speed, memory usage,

Benchmarking experiments were done using the JLoop (Java)

multiprocessing, etc. of visualization software. The

and Loop (C/C++) visualization software for animation of 2D

performance of JLoop and QtLoop/Loop was compared for

datasets. Preliminary results for six different hardware

interactive animation (display) and numerical computation.

platforms showed that compared to the C/C++ version, the

Java implementation was four to five times slower for graphics RESULTS

image animation, used on average 2.5 to three times more Graphics rendering (and reading) in JLoop of a large data set

memory, and was up to 3.6 times slower using a pure numerical such as Hurricane George GOES imagery, 3600 x 3600 x 2 4

benchmark that simulated image pixel operations. Such bits/frame x 500 frames or 39MB each, was 4 to 5 times slower

benchmarks should be used to improve the Java VM than Qt/Loop. Animation of an image sequence can be

performance across platforms, similar to JIT-compiler design. implemented utilizing the MediaTracker class with one or

General Terms: Performance, Languages, Experimentation many portable (green) threads. Ensuring smooth flicker-free

animation and loading is a challenging task in JLoop. Memory

Keywords: Cross-platform visualization, benchmarking, high

usage was up to five times higher. Computational performance

performance, Java, C/C++, Java advanced imaging

using Java 2 v1.4 for a hundred million evaluations of

SUMMARY different quadratics was up to 2.8 times slower than compiled

Scientific visualization applications that require interactivity C, both with optimization flags (see Table); SGI IRIX used

pose significant software challenges in algorithm Java 1.2. Penalty for disabling JIT was 44.41 sec for Linux or

development and optimization for handling large datasets, 16X slower than C. Other limitations include 2GB-program

efficient cache management, and parallel execution, have memory limit (except on Solaris), and the present lack of

traditionally been developed in C or C++. Software developed portability of Java Advanced Imaging (JAI) classes.

in Java offers compelling advantages such as platform

WinP3 WinP4 Linux SUN Alpha SGI

independence, architecture neutrality, built-in support for

11.03 7.19 4.36 28.42 18.98 36.06

Java

multithreading and distributed programming, automatic

8.55 2.59 2.78 23.23 7.17 14.38

C

garbage collection, a simpler object-oriented framework and

Ratio 1.290 2.781 1.566 1.223 2.647 2.508

compiled or interpreted mode of execution. However, the

performance differences between Java and C/C++ Comparison of execution speeds for polynomial evaluation

implementations of the same visualization functionality needs

REFERENCES

more thorough investigation in order to quantify software

[1] I. H. Kazi, H. H. Chen, B. Stanley, and D. J. Lilja,

engineering tradeoffs for future projects and to improve the

Techniques for obtaining high performance in Java

capabilities of the Java Virtual Machine (VM). The relative

programs, ACM Computing Surveys, 32(3):213-240, 2000.

performance difference of the SUN JDK v1.0 or 1.1 interpreter

[2] Austin, C. and M. Pawlan, Advanced Programming for the

was 10 to 50 times slower than compiled C code on the same

Java 2 Platform, Chap. 8, Addison-Wesley Professional, 2002.

target processor [1]. This deficiency led to the development of

[3] A.F. Hasler, K. Palaniappan, M. Manyin, J. Dodge, A high

Just-In-Time (JIT) compiler technology, for increasing

performance interactive image spreadsheet (IISS), C omputers

performance so that execution times became only 1.5 to 4

in Physics, Vol. 8, No. 4, 1994, pp. 325-342.

times slower [1]. A JIT-compiler is a code generator that

[4] K. Palaniappan, A. F. Hasler, J. Fraser, M. Manyin Network-

converts Java bytecode into reusable native machine code and

based visualization using the Distributed Image SpreadSheet

is now a standard tool in the current SUN Java 2 SDK Standard

(DISS), 17th Int. Conf. on Interactive Information and

Edition v1.4 [2]. The IISS [3] and DISS [4] visualization

Processing Systems (IIPS), 81st AMS, American Meteorological

Society, Albuquerque, NM, 2001, pp. 399-403.

Copyright is held by the author/owner(s).

JGI'02, November 3 5, 2002, Seattle, Washington, USA.

ACM 1-58113-599-8/02/0011.

Society, Albuquerque, NM, 2001, pp. 399-403.

Copyright is held by the author/owner(s).

Contact this candidate