Java 5 is VERY slow on Linux Fedora core3

Question

Why Java is 10-15 times slower on Fedora Linux, in comparising to Windows XP?
I have two identical PC (dual Pentuim3 1Gh, 1.5Gig memory).
Java version - JDK 1.5.0_03
For testing, I run simple java program:
/// Start LoadTest.Java
import java.util.*;
public class LoadTest {
public static void main(String[] args)
Date startDate = new Date();
int repeat = 1000;
HashMap map = new HashMap();
for(int i=0; i<repeat; i++) {
//System.out.println(i);
ArrayList list = new ArrayList();
long lval = (new Date()).getTime() / startDate.getTime() + ((new Date()).getTime() * startDate.getTime()) - startDate.getTime();
String sval = Long.toString(lval);
Long LValue = Long.valueOf( sval ).longValue();
map.put(LValue, sval);
String keyset = map.keySet().toString();
list.add(0,keyset);
list.addAll(map.values());
for (Iterator iter = map.keySet().iterator(); iter.hasNext(); ) {
Long lv = (Long) iter.next();
String sv = (String) map.get(lv);
list.add(sv);
keyset = map.keySet().toString();
list.add(0,keyset);
Date endDate = new Date();
String message = " LoadTest: " + repeat + " repetitions. Total Time: " + (endDate.getTime() - startDate.getTime());
System.out.println(message);
/// End LoadTest.Java
It takes 300ms to run it on Windows and 3900ms - on Linux
Can somebody help me?
Shall I consider to use Windows on the web-server?
Thanks.

Answer 1

I know, it's just dirty, ugly test, but WHY WinXP handles it better?
Actually, I did this best after I get som test results for my application.
I't uses JBoss4.1/Tomcat5 on JDK 1.5_03. I used JMeter to test it and get some strange results - to test throughput rate on plain JSP pages, like login page (without any EJB) , Linux performs 10 times slower (for 100 requests, 24/min on Linux and 240/min on Windows). To test EJB, which retrieving datasets from database, the difference is about 3 times. Linux - FedoraCore 3 (2.6.11)
That's why I started testing Java itself. BTW, if make this test a little "pretty" - no memory allocations in loop, less using arraylists and maps, the performance difference about 3 times or even less, if to put Thread.sleep(1) in the loop.
Now, I have to figure out, which paltform to use for our web-application, but I'm confused.
BTW, which app.server did you use? May be JBoss is not the best solution? The application should handle about 200 users, most of them almost simultaneously.
Thanks