I am actually running gentoo linux with glibc 2.3.1 which does not have the
LD_ASSUME_KERNEL option. From what I know, this forces glibc 2.1.3 (on
Redhat) to be used. However gentoo does not support this from what I can
tell. The system breaks when this variable is used.
On a performance note, I ran some more benchmarks for threads on both
blackdown 1.4.1 and IBM 1.3.1 and on a 2.5.47 kernel and the results are
below. From what I know both blackdown and IBM use native threads. As the
results show, (for essentially a null thread) as the number of threads
increase the ibm implementation is better and better. The main reason I
believe it is slower for the smaller number of threads is the page fault
penalty. This is of course just curiousity for now, and I can't really
imagine any application where you create and destroy 100,000 threads leave
alone 1,000,000 threads in such a short amount of time.
Naren
-----------------------------benchmark --------------------------------
class Loop implements Runnable {
public static void main(String[] args) {
for (int t = 0; t < Integer.parseInt(args[0]); t++)
new Thread(new Loop()).start();
}
public void run() {
int i;
for (i = 0; i < 1; i++)
for (int j = 0; j < 1; j++);
}
}
----------------------------- Black down 1.4.1
time java -server Loop 1
0.199u 0.027s 0:00.23 91.3% 0+0k 0+0io 1921pf+0w
time java -server Loop 10
0.196u 0.030s 0:00.22 100.0% 0+0k 0+0io 1921pf+0w
time java -server Loop 100
0.224u 0.036s 0:00.26 96.1% 0+0k 0+0io 1921pf+0w
time java -server Loop 1000
0.382u 0.218s 0:00.60 98.3% 0+0k 0+0io 1930pf+0w
time java -server Loop 10000
2.198u 2.014s 0:03.67 114.4% 0+0k 0+0io 2067pf+0w
time java -server Loop 100000
14.613u 17.821s 0:30.44 106.5% 0+0k 0+0io 2074pf+0w
time java -server Loop 1000000
137.510u 181.709s 5:04.01 104.9% 0+0k 0+0io 2073pf+0w
---------------------------- IBM 1.3.1
time /opt/ibm-jdk-1.3.1/bin/java Loop 1
0.226u 0.050s 0:00.30 90.0% 0+0k 0+0io 2972pf+0w
time /opt/ibm-jdk-1.3.1/bin/java Loop 10
0.245u 0.039s 0:00.29 93.1% 0+0k 0+0io 2972pf+0w
time /opt/ibm-jdk-1.3.1/bin/java Loop 100
0.227u 0.064s 0:00.31 90.3% 0+0k 0+0io 2972pf+0w
time /opt/ibm-jdk-1.3.1/bin/java Loop 1000
0.356u 0.076s 0:00.64 65.6% 0+0k 0+0io 2978pf+0w
time /opt/ibm-jdk-1.3.1/bin/java Loop 10000
0.817u 0.290s 0:02.95 37.2% 0+0k 0+0io 3003pf+0w
time /opt/ibm-jdk-1.3.1/bin/java Loop 100000
3.752u 2.242s 0:22.71 26.3% 0+0k 0+0io 3003pf+0w
time /opt/ibm-jdk-1.3.1/bin/java Loop 1000000
36.428u 21.904s 3:53.08 25.0% 0+0k 0+0io 3003pf+0w
On Wednesday 20 November 2002 02:16 am, Marco Trevisan wrote:
> Hi,
>
> Remember that IBM-jvm needs the following environment variable:
>
> LD_ASSUME_KERNEL=2.2.5
>
> You could try this and see if page faults continue arising.
>
> Regards,
> Marco Trevisan
>
> Narendra Sankar wrote:
> >hi
> >
> >I ran a very simple thread creation benchmark on various vms to find out
> > how useful my two processors were for java threads and this is what I get
> > -
>
> [...]
>
>
> ----------------------------------------------------------------------
> To UNSUBSCRIBE, email to [EMAIL PROTECTED]
> with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
