Re: [rvm-core] [rvm-commits] SF.net SVN: jikesrvm: [14159] rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/ FreeListPageResource.java

View: New views
3 Messages — Rating Filter:   Alert me  

Parent Message unknown Re: [rvm-core] [rvm-commits] SF.net SVN: jikesrvm: [14159] rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/ FreeListPageResource.java

by Ian Rogers :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

Hi Steve,

just a comment, this patch has given a fairly large regression on the
semi-space allocation rate (~50%):

http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32b.anu.edu.au/sanity/3600/FastAdaptiveSemiSpace/Performance/gctest/FixedLive/SemiSpace%20AllocationRate.large.png

and also the mark sweep allocation rate (~12%):

http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32b.anu.edu.au/sanity/3600/FastAdaptiveMarkSweep/Performance/gctest/FixedLive/MarkSweep%20AllocationRate.large.png

presumably this was to be expected.

Regards,
Ian

steveb-oss@... wrote:

> Revision: 14159
>           http://jikesrvm.svn.sourceforge.net/jikesrvm/?rev=14159&view=rev
> Author:   steveb-oss
> Date:     2008-04-23 20:00:32 -0700 (Wed, 23 Apr 2008)
>
> Log Message:
> -----------
> Bug fix. Correctly identify new chunks when using discontiguous spaces.
>
> Modified Paths:
> --------------
>     rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/FreeListPageResource.java
>
> Modified: rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/FreeListPageResource.java
> ===================================================================
> --- rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/FreeListPageResource.java 2008-04-23 22:02:47 UTC (rev 14158)
> +++ rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/FreeListPageResource.java 2008-04-24 03:00:32 UTC (rev 14159)
> @@ -146,8 +146,10 @@
>      lock();
>      boolean newChunk = false;
>      int pageOffset = freeList.alloc(pages);
> -    if (pageOffset == GenericFreeList.FAILURE && !contiguous)
> +    if (pageOffset == GenericFreeList.FAILURE && !contiguous) {
>        pageOffset = allocateContiguousChunks(pages);
> +      newChunk = true;
> +    }
>      if (pageOffset == -1) {
>        unlock();
>        return Address.zero();
>
>
> This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
> Don't miss this year's exciting event. There's still time to save $100.
> Use priority code J8TL2D2.
> http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
> _______________________________________________
> Jikesrvm-commits mailing list
> Jikesrvm-commits@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-commits
>  


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Jikesrvm-core mailing list
Jikesrvm-core@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-core

Re: [rvm-core] [rvm-commits] SF.net SVN: jikesrvm: [14159] rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/ FreeListPageResource.java

by Steve Blackburn :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

Thanks for spotting that Ian.   It was not expected.  I'll chase it  
down.

Ugh.  We desperately need something that consolidates the sea of  
regression data and highlights the most important issues.   This one  
flew under the radar.

--Steve

On 29/04/2008, at 9:13 PM, Ian Rogers wrote:

> Hi Steve,
>
> just a comment, this patch has given a fairly large regression on the
> semi-space allocation rate (~50%):
>
> http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32b.anu.edu.au/sanity/3600/FastAdaptiveSemiSpace/Performance/gctest/FixedLive/SemiSpace%20AllocationRate.large.png
>
> and also the mark sweep allocation rate (~12%):
>
> http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32b.anu.edu.au/sanity/3600/FastAdaptiveMarkSweep/Performance/gctest/FixedLive/MarkSweep%20AllocationRate.large.png
>
> presumably this was to be expected.
>
> Regards,
> Ian
>
> steveb-oss@... wrote:
>> Revision: 14159
>>          http://jikesrvm.svn.sourceforge.net/jikesrvm/?rev=14159&view=rev
>> Author:   steveb-oss
>> Date:     2008-04-23 20:00:32 -0700 (Wed, 23 Apr 2008)
>>
>> Log Message:
>> -----------
>> Bug fix. Correctly identify new chunks when using discontiguous  
>> spaces.
>>
>> Modified Paths:
>> --------------
>>    rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
>> FreeListPageResource.java
>>
>> Modified: rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
>> FreeListPageResource.java
>> ===================================================================
>> --- rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
>> FreeListPageResource.java 2008-04-23 22:02:47 UTC (rev 14158)
>> +++ rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
>> FreeListPageResource.java 2008-04-24 03:00:32 UTC (rev 14159)
>> @@ -146,8 +146,10 @@
>>     lock();
>>     boolean newChunk = false;
>>     int pageOffset = freeList.alloc(pages);
>> -    if (pageOffset == GenericFreeList.FAILURE && !contiguous)
>> +    if (pageOffset == GenericFreeList.FAILURE && !contiguous) {
>>       pageOffset = allocateContiguousChunks(pages);
>> +      newChunk = true;
>> +    }
>>     if (pageOffset == -1) {
>>       unlock();
>>       return Address.zero();
>>
>>
>> This was sent by the SourceForge.net collaborative development  
>> platform, the world's largest Open Source development site.
>>
>> -------------------------------------------------------------------------
>> This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
>> Don't miss this year's exciting event. There's still time to save  
>> $100.
>> Use priority code J8TL2D2.
>> http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
>> _______________________________________________
>> Jikesrvm-commits mailing list
>> Jikesrvm-commits@...
>> https://lists.sourceforge.net/lists/listinfo/jikesrvm-commits
>>
>
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
> Don't miss this year's exciting event. There's still time to save  
> $100.
> Use priority code J8TL2D2.
> http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
> _______________________________________________
> Jikesrvm-core mailing list
> Jikesrvm-core@...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-core


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Jikesrvm-core mailing list
Jikesrvm-core@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-core

Re: [rvm-core] [rvm-commits] SF.net SVN: jikesrvm: [14159] rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/ FreeListPageResource.java

by Steve Blackburn :: Rate this Message:

Reply to Author | View Threaded | Show Only this Message

I have created a JIRA to deal with this: http://jira.codehaus.org/browse/RVM-470

--Steve

On 30/04/2008, at 6:10 AM, Steve Blackburn wrote:
Thanks for spotting that Ian.   It was not expected.  I'll chase it  
down.

Ugh.  We desperately need something that consolidates the sea of  
regression data and highlights the most important issues.   This one  
flew under the radar.

--Steve

On 29/04/2008, at 9:13 PM, Ian Rogers wrote:

Hi Steve,

just a comment, this patch has given a fairly large regression on the
semi-space allocation rate (~50%):

http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32b.anu.edu.au/sanity/3600/FastAdaptiveSemiSpace/Performance/gctest/FixedLive/SemiSpace%20AllocationRate.large.png

and also the mark sweep allocation rate (~12%):

http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32b.anu.edu.au/sanity/3600/FastAdaptiveMarkSweep/Performance/gctest/FixedLive/MarkSweep%20AllocationRate.large.png

presumably this was to be expected.

Regards,
Ian

steveb-oss@... wrote:
Revision: 14159
        http://jikesrvm.svn.sourceforge.net/jikesrvm/?rev=14159&view=rev
Author:   steveb-oss
Date:     2008-04-23 20:00:32 -0700 (Wed, 23 Apr 2008)

Log Message:
-----------
Bug fix. Correctly identify new chunks when using discontiguous  
spaces.

Modified Paths:
--------------
  rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
FreeListPageResource.java

Modified: rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
FreeListPageResource.java
===================================================================
--- rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
FreeListPageResource.java 2008-04-23 22:02:47 UTC (rev 14158)
+++ rvmroot/trunk/MMTk/src/org/mmtk/utility/heap/
FreeListPageResource.java 2008-04-24 03:00:32 UTC (rev 14159)
@@ -146,8 +146,10 @@
   lock();
   boolean newChunk = false;
   int pageOffset = freeList.alloc(pages);
-    if (pageOffset == GenericFreeList.FAILURE && !contiguous)
+    if (pageOffset == GenericFreeList.FAILURE && !contiguous) {
     pageOffset = allocateContiguousChunks(pages);
+      newChunk = true;
+    }
   if (pageOffset == -1) {
     unlock();
     return Address.zero();


This was sent by the SourceForge.net collaborative development  
platform, the world's largest Open Source development site.

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save  
$100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Jikesrvm-commits mailing list
Jikesrvm-commits@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-commits



-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save  
$100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Jikesrvm-core mailing list
Jikesrvm-core@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-core


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Jikesrvm-core mailing list
Jikesrvm-core@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-core


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Jikesrvm-core mailing list
Jikesrvm-core@...
https://lists.sourceforge.net/lists/listinfo/jikesrvm-core
LightInTheBox - Buy quality products at wholesale price