harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Aleksey Shipilev" <aleksey.shipi...@gmail.com>
Subject Re: [drlvm] performance diminishing while executing the test
Date Wed, 28 Nov 2007 16:29:11 GMT
Andrey,

I tried to find the test you mentioned:

shade@mylovelybox /c/users/shade/Harmony/trunk
$ find | grep -i WeakReferenceandThreadTest

shade@mylovelybox /c/users/shade/Harmony/trunk
$ find | grep -i WeakReference | grep -v svnbase
./working_classlib/build/classes/java/lang/ref/WeakReference.class
./working_classlib/modules/luni/src/test/api/common/tests/api/java/lang/ref/WeakReferenceTest.java
./working_classlib/modules/luni-kernel/src/main/java/java/lang/ref/WeakReference.java
./working_vm/vm/tests/smoke/gc/WeakReferenceTest.java
./working_vm/vm/vmcore/src/kernel_classes/javasrc/java/lang/ref/WeakReference.java

I can't see the test you've mentioned... Can you point to exact
location of the test?

Thanks,
Aleksey,
ESSD, Intel.

On Nov 28, 2007 7:11 PM, Andrey Yakushev <andrey.yakushev@gmail.com> wrote:
> I just found curious behavior of DRLVM. When running
> WeakReferenceandThreadTest from the reliability suite during different
> time period I noted that DRLVM performance gradually diminishes during
> the functioning. The results:
>
> 5 min test – 532 test iterations
> 10 min – 809 iterations
> 15 min – 983
> 20 min – 1150
> 30 min – 1369
> 60 min – 1779
>
> So during the second half of hour in the one hour test run DRLVM
> executes less iterations than single 5 min test run.
> I think these results should be investigated. I suspect this in weak
> reference related peculiarities, but not sure.
>
> This is also an explanation for reliability test failing by timeout
> which we have periodically. Our algorithm estimates overall test
> duration by linear interpolation from the first test run. I'll try to
> fix this in reliability harness.
>
> --
> Thanks,
> Andrey
>

Mime
View raw message