LISTSERV mailing list manager LISTSERV 16.0

Help for ECJ-INTEREST-L Archives


ECJ-INTEREST-L Archives

ECJ-INTEREST-L Archives


ECJ-INTEREST-L@LISTSERV.GMU.EDU


View:

Message:

[

First

|

Previous

|

Next

|

Last

]

By Topic:

[

First

|

Previous

|

Next

|

Last

]

By Author:

[

First

|

Previous

|

Next

|

Last

]

Font:

Proportional Font

LISTSERV Archives

LISTSERV Archives

ECJ-INTEREST-L Home

ECJ-INTEREST-L Home

ECJ-INTEREST-L  February 2006

ECJ-INTEREST-L February 2006

Subject:

Garbage Collection and a Simple 34% Performance Tweak

From:

Chris Ellingwood <[log in to unmask]>

Reply-To:

ECJ Evolutionary Computation Toolkit <[log in to unmask]>

Date:

Wed, 1 Feb 2006 10:52:57 -0500

Content-Type:

text/plain

Parts/Attachments:

Parts/Attachments

text/plain (131 lines)

The short version of the story - By setting gc=false and adding the 
following parameter to the VM command line" -XX:+AggressiveHeap, ECJ used 
28% less computational time. Same seed. 
  The performance increase is a result of instructing the JVM to use a 
more efficient garbage collection algorithm.
  Note, using this switch uses memory more aggressively, so you may want 
to read on.

  The long version of the story. 

  I have been using ECJ for the last year and in the past few months 
became interested in increasing performance. To that end, I added timers 
to each of the major operations of ECJ. Here's what a typical report my 
look like:

> Java Default GC Parameters
>
> Run Time:93.94 -> 100.00%
> Setup:3.98 -> 4.23%
> Evaluate:17.94 -> 19.09%
> PostEval Stats:4.32 -> 4.60%
> Breed:67.02 -> 71.34%
> Final Stats:0.90 -> 0.96%

  First number is seconds and second number is percent of CPU time. 
  I was surprised to see that breeding consumed such a large proportion of 
the CPU consumption - 71%. Timers were added to specific functions inside 
of breeding:

> Run Time:93.94 -> 100.00%
> Setup:3.98 -> 4.23%
> Evaluate:17.94 -> 19.09%
> PostEval Stats:4.32 -> 4.60%
> Breed:67.02 -> 71.34%
> --PrepareToProduce:7.47 -> 7.95%
> --Produce:59.38 -> 63.21%
> ----Produce_BreedingSources:0.78 -> 0.83%
> ----Produce_Produce:57.95 -> 61.68%
> ------CrossOverPipeline_Breed6:33.62 -> 35.79%
> -------CrossOverPipeline_Breed7_childAllocate:8.02 -> 8.54%
> Final Stats:0.90 -> 0.96%

One line of code “obj.children = new GPNode[children.length]” was found to 
consume 8.54% of the CPU time. This line of code allocates memory from the 
heap for tree children - a very common operation in breeding.

A first tack against this consumer was to think that dynamic object 
allocation was eating processor time. This could be cured by statically 
defining a collection of nodes that the GP could recycle through without 
JVM memory allocation/de-allocation. This was briefly explored until 
realizing that creating/maintaining such a collection would require 
extensive code changes.

A second and ultimately successful tack was to consider the de-allocation 
side. Was the JVM’s garbage collection consuming the processor time? 
Garbage collection is performed by the JVM without needing to be 
programmed directly – one of the beauties of Java – no memory leaks. In 
most programs, this does not introduce performance degradation. However, 
most programs don’t allocate millions of objects per second as ECJ. 

While Sun Java does offer some access to performing explicit garbage 
collection in code, doing this in ECJ only introduced additional 
performance degradation. The other way to control GC is through JVM 
command line parameters. After a bit of internet research, the following 
command GC/Heap parameters were discovered and tested.

> 75000 individuls x 5gen
> Command Line Switch Total Run Time
> --none--                    11.5s
> -XX:+UseParallelGC          13.7s
> -XX:+UseConcMarkSweepGC     16.5s
> -Xincgc                     16.7s
> -XX:+UseAdaptiveSizePolicy  11.6s
> -XX:+AggressiveHeap          7.2s

After determining the best command line switch, I tested combinations of 
ECJ's garabge collection and the switch. 

> 75000 inds x 50gen – testing switch with ECJ code GC
> Command Line Switch ECJ GC Total Run Time
> --none--              none      93.9s
> -XX:+AggressiveHeap   none      62.0s
> --none--              normal      100.1s
> -XX:+AggressiveHeap   normal      78.8s
> -none                 aggreesive  111.2s
> -XX:+AggressiveHeap   aggressive   91.0s

As can be observed from the above results, using -XX:+AggressiveHeap and 
turning off ECJ GC resulted in the much fastred run times by far. 

A note on memory usage. -XX:+AggressiveHeap is just that - aggressive. If 
you get an "Out of Heap Memory" error when using this switch, then you 
will need to setthe VM parameter Xmx???m. This parameter sets the maximum 
heap size and units are megabytes. The smallest value I ever use is: 
Xmx75m. I'm running symbolic regression with average trees sizes of 18. My 
rule of thumb is to allocate 2M of memory for every 1,000 individuals. For 
example, a run of 200,000 individuals, I will set the parameter to 
Xmx400m. Your needs may be more or less depending on your tree/population 
sizes. There's a balance here. It needs to be set high enough so there's 
enough heap space. It needs to be set low enough as to not aggressively 
consume so much memory that you start doing disk page swaps, especially if 
you're running multiple simultaneous runs. Again, this might not be an 
issue for smaller populations, non-paralell runs or if you have vast 
amounts of memory. 

For more information on the -XX:+AggressiveHeap switch and other heap 
related switches:
http://java.sun.com/j2se/1.4.2/reference/whitepapers/

  I'd love to hear back from any of you who experiment with this. 

  Also, if there's popular demand, I can include the Performance 
Monitoring Class that I've created.

  best initentions, Chris Ellingwood

(University of Vermont, Botany)

-----------

Configuration I:
Sun Java: 1.5.0_05
Dual Processor Xeon 3.6 GHZ 
Windows XP - 64 bit
ECJ 11

Configuration II:
Sun Java: 1.5.0_05
Pentium 4 - 2.4 GHZ 
Windows XP
ECJ 11

Top of Message | Previous Page | Permalink

Advanced Options


Options

Log In

Log In

Get Password

Get Password


Search Archives

Search Archives


Subscribe or Unsubscribe

Subscribe or Unsubscribe


Archives

June 2019
April 2019
March 2019
January 2019
December 2018
November 2018
October 2018
July 2018
May 2018
January 2018
December 2017
November 2017
October 2017
September 2017
August 2017
July 2017
June 2017
May 2017
April 2017
March 2017
February 2017
January 2017
December 2016
November 2016
October 2016
September 2016
August 2016
July 2016
June 2016
May 2016
April 2016
March 2016
February 2016
January 2016
December 2015
November 2015
October 2015
September 2015
August 2015
July 2015
June 2015
May 2015
April 2015
March 2015
February 2015
January 2015
December 2014
November 2014
September 2014
August 2014
July 2014
June 2014
May 2014
April 2014
March 2014
February 2014
January 2014
December 2013
November 2013
October 2013
September 2013
August 2013
July 2013
June 2013
May 2013
April 2013
March 2013
February 2013
January 2013
December 2012
November 2012
October 2012
September 2012
July 2012
June 2012
May 2012
April 2012
March 2012
February 2012
January 2012
December 2011
November 2011
October 2011
September 2011
August 2011
July 2011
June 2011
May 2011
April 2011
March 2011
February 2011
January 2011
December 2010
November 2010
October 2010
September 2010
August 2010
July 2010
June 2010
May 2010
April 2010
March 2010
February 2010
January 2010
December 2009
November 2009
October 2009
September 2009
August 2009
July 2009
June 2009
May 2009
April 2009
March 2009
February 2009
January 2009
December 2008
November 2008
October 2008
September 2008
August 2008
July 2008
June 2008
May 2008
April 2008
March 2008
February 2008
January 2008
December 2007
November 2007
October 2007
September 2007
August 2007
July 2007
June 2007
May 2007
April 2007
March 2007
February 2007
January 2007
December 2006
November 2006
October 2006
September 2006
August 2006
July 2006
June 2006
May 2006
April 2006
March 2006
February 2006
January 2006
December 2005
November 2005
October 2005
September 2005
August 2005
July 2005
June 2005
May 2005
April 2005
March 2005
February 2005
January 2005
December 2004
November 2004
September 2004
August 2004
July 2004
June 2004
May 2004
April 2004
March 2004

ATOM RSS1 RSS2



LISTSERV.GMU.EDU

CataList Email List Search Powered by the LISTSERV Email List Manager