You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@commons.apache.org by "Gilles (Created) (JIRA)" <ji...@apache.org> on 2012/02/09 16:37:59 UTC
[jira] [Created] (MATH-740) Some "FastMath" functions are slow
Some "FastMath" functions are slow
----------------------------------
Key: MATH-740
URL: https://issues.apache.org/jira/browse/MATH-740
Project: Commons Math
Issue Type: Wish
Reporter: Gilles
Priority: Minor
Fix For: 3.1
>From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
* log10 (46%, 36%)
* log1p (68%, 112%)
* tan (11%, 61%)
* atan (26%, 125%)
* atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Leandro Ariel Pezzente (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13222712#comment-13222712 ]
Leandro Ariel Pezzente commented on MATH-740:
---------------------------------------------
[~ijuma] Have you ran any of these benchmarking tests on Oracle JRockit VM to account for any extra optimization in this particular VM ?
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Gilles (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13428700#comment-13428700 ]
Gilles commented on MATH-740:
-----------------------------
Revision 1369506 contains a few changes that improve the performance of "log" by a small amount.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Gilles (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13428709#comment-13428709 ]
Gilles commented on MATH-740:
-----------------------------
{noformat}
log and log10 (calls per timed block: 1000000, timed blocks: 100, time unit: ms)
name time/call std error total time ratio difference
Math log(x) 3.61083035e-05 2.28393527e-06 3.6108e+03 1.0000e+00 0.00000000e+00
Math log10(x) 3.66085799e-05 3.22000007e-06 3.6609e+03 1.0139e+00 5.00276380e+01
StrictMath log(x) 9.51341413e-05 2.66850659e-06 9.5134e+03 2.6347e+00 5.90258378e+03
StrictMath log10(x) 1.23497329e-04 3.77763096e-06 1.2350e+04 3.4202e+00 8.73890257e+03
FastMath log(x) 6.35533555e-05 5.03781330e-06 6.3553e+03 1.7601e+00 2.74450520e+03
FastMath log10(x) 1.66252138e-04 6.30307785e-06 1.6625e+04 4.6043e+00 1.30143835e+04
FastMath log(10, x) 1.17000075e-04 3.14090930e-06 1.1700e+04 3.2403e+00 8.08917711e+03
{noformat}
This table shows that
# FastMath's log is about 1.7 times slower than Math's log
# FastMath's log10 is about 4.5 times slower than Math's log10
# FastMath's log(10, x), which calls log twice, is about 3.2 times slower than Math's log10
Small note: the performance difference between Math's log and log10 is not significant (in some runs, log is faster, in others log10 is faster).
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Leandro Ariel Pezzente (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13222963#comment-13222963 ]
Leandro Ariel Pezzente commented on MATH-740:
---------------------------------------------
I really didnt knew that fact. That would greaty optimize Java VM.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Thomas Neidhart (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13253415#comment-13253415 ]
Thomas Neidhart commented on MATH-740:
--------------------------------------
I am a bit puzzled, does anybody have an idea why Math is faster than StrictMath, although internally, Math.log10 just calls StrictMath.log10?
{noformat}
Name StrictMath FastMath Math Runs=10000000 Java 1.6.0_20 (1.6.0_20-b02) Java HotSpot(TM) Server VM (16.3-b01)
log10 98 1.0 161 1.6377 23 0.2427
{noformat}
>From Math.java:
{noformat}
public static double log10(double a) {
return StrictMath.log10(a); // default impl. delegates to StrictMath
}
{noformat}
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Ismael Juma (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13222248#comment-13222248 ]
Ismael Juma commented on MATH-740:
----------------------------------
I understand. My first comment in this was posted before I saw your reply in MATH-580. The second comment was simply stating that the issue may have been fixed. I haven't had a chance to verify yet.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Gilles (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13221683#comment-13221683 ]
Gilles commented on MATH-740:
-----------------------------
Hi.
I'm sorry but I don't understand what your comment aims at. My previous post was asking whether _you_ could have a look at _our_ benchmarking utility, in order to check that the methodology is appropriate and whether this micro-benchmarking code gives results in accordance to your tests.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Ismael Juma (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13222782#comment-13222782 ]
Ismael Juma commented on MATH-740:
----------------------------------
Leandro, I haven't. Also, not sure the value of doing so as Oracle has stated that JRockit will be merged into HotSpot. And some of that has already happened.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Thomas Neidhart (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13253448#comment-13253448 ]
Thomas Neidhart commented on MATH-740:
--------------------------------------
ah ok thanks. So the call in Math.log10 is just there to make the javac happy while the jvm will call the intrinsic methods instead.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Ismael Juma (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13253452#comment-13253452 ]
Ismael Juma commented on MATH-740:
----------------------------------
That is true for HotSpot indeed. For other JVMs that use the OpenJDK standard library, they may or may not have similar mechanisms.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Ismael Juma (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13253421#comment-13253421 ]
Ismael Juma commented on MATH-740:
----------------------------------
Thomas, you cannot trust what you see in java.lang.Math. Many of those methods are HotSpot intrinsics and consist of optimised assembly code. Examples are sin, cos, tan, exp, log, sqrt, atan2 and so on.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Ismael Juma (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13211964#comment-13211964 ]
Ismael Juma commented on MATH-740:
----------------------------------
Regarding the last comment, Gilles pointed out in MATH-580 that there have been updates to the benchmarking code since the blog entry was written:
https://issues.apache.org/jira/browse/MATH-580?focusedCommentId=13128547&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13128547
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Updated] (MATH-740) Some "FastMath" functions are slow
Posted by "Gilles (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Gilles updated MATH-740:
------------------------
Fix Version/s: (was: 3.1)
3.2
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.2
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Gilles (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13204607#comment-13204607 ]
Gilles commented on MATH-740:
-----------------------------
I should note that the above numbers seem to vary wildly from run to run (my machine is fairly loaded at the moment) but the trend is there: Always too slow.
Looking at the code, it seems that there are potential source of inefficiencies (like the missing "final" keywords which I've added, in a recent commit for "atan") or accessing array locations (instead of storing the value in a local variable), but each of those changes is likely to not demonstrate a significant improvement. :(
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] [Commented] (MATH-740) Some "FastMath" functions are slow
Posted by "Ismael Juma (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/MATH-740?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13211952#comment-13211952 ]
Ismael Juma commented on MATH-740:
----------------------------------
Also see:
http://blog.juma.me.uk/2011/02/23/performance-of-fastmath-from-commons-math
It includes some fixes to the benchmarking code.
> Some "FastMath" functions are slow
> ----------------------------------
>
> Key: MATH-740
> URL: https://issues.apache.org/jira/browse/MATH-740
> Project: Commons Math
> Issue Type: Wish
> Reporter: Gilles
> Priority: Minor
> Fix For: 3.1
>
>
> From the two benchmarks we currently have in "FastMathTestPerfomance", we have that the following functions are much slower in "FastMath" than in either "Math" or "StrictMath" (the performance *loss*, for each of the benchmarks, is given in parentheses):
> * log10 (46%, 36%)
> * log1p (68%, 112%)
> * tan (11%, 61%)
> * atan (26%, 125%)
> * atan2 (44%, 40%)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira