Updated Java vs. C++

Updated to bring to the current state of the art in 2003 results of Java vs. C++

Comparison using generic bubble sort

Source code

I have copied from the original website and made available here the source code with following modifications:

• repeated sorting 10 times to increase timing from like 2secs to 20sec ...
• in C++ used time.h to automate timings (second resolution is find for this kind of tests)
• in Java used System.currrentTime() to mesaure time
• ported C++ version to use #include <vector> and added using namespace std;
• converted C++ code to use printfs intead of cout ...
• minor changes to class namings, in Java moved Lib as inner static class

I have also added two handy script ot simplify compiling code with VC++ (vc_compile.bat) and with JAVA (java_compile.bat).

C++ version

Java version

Updated Results

I have run tests on 850 MHz Pentium III PC running Windows 2000 SP3: (to compile C++ code i used command line cl /O2 and for vector and STL added /GX option) detailed log of test execution is available.

Version	Execution Environment	Execution Time
C++ pointer	MS Visual C++ 6.0 SP5	19s
C++ object	MS Visual C++ 6.0 SP5	11s
C++ vector	MS Visual C++ 6.0 SP5	13s
C++ STL	MS Visual C++ 6.0 SP5	12s
Java array	HotSpot Client VM (build 1.4.1_01-b01, mixed mode)	103s
	HotSpot(TM) Server VM (build 1.4.1_01-b01, mixed mode)	89s
Java Vector	HotSpot Client VM (build 1.4.1_01-b01, mixed mode)	355s
	HotSpot(TM) Server VM (build 1.4.1_01-b01, mixed mode)	241s
Java ArrayList	HotSpot Client VM (build 1.4.1_01-b01, mixed mode)	328s
	HotSpot(TM) Server VM (build 1.4.1_01-b01, mixed mode)	114s

I have also received from Dave Marquar interesting results (see that client Hotspot Vector is faster than ArrayList) for running Java tests on an Athlon 2.0+ running Windows XP with JDK 1.4.1:

Version	Execution Environment	Execution Time
Java array	HotSpot Client VM (build 1.4.1_01-b01, mixed mode)	50s
	HotSpot(TM) Server VM (build 1.4.1_01-b01, mixed mode)	46s
Java Vector	HotSpot Client VM (build 1.4.1_01-b01, mixed mode)	144s
	HotSpot(TM) Server VM (build 1.4.1_01-b01, mixed mode)	105s
Java ArrayList	HotSpot Client VM (build 1.4.1_01-b01, mixed mode)	149s
	HotSpot(TM) Server VM (build 1.4.1_01-b01, mixed mode)	42s

There are few interesting observations that one an make. First: using arrays instead of Vector/ArrayList in very computational intensive applications may make sense. Secondly the difference between Vector and ArrayList is no longer significant in moden JVMs. Finally Java Hotspot server is very good optimizer and using ArrayList in such setup gives performance close to native java array and this is indeed impressive result!