From: ramalingam rajaraman [rraman64@gmail.com] Sent: Thursday, 25 December 2008 00:31 To: Nielsen Ole Cc: josephwinston@gmail.com; Wilson Ross Subject: Re: anuga in 64 bit vs 32 bit [SEC=UNCLASSIFIED] Dear Ole, MERRY CHRISTMAS. I finally made a small test run with our realistic data to understand the differences. before going into 32bit vs 64bit, I wanted to compare how they behave in 32bit itself with different OS. I found xp and vista are giving identical results irrespective of anuga version (SVN vs 5953) both at land and at sea locations. Ubuntu results for sea location is exactly same as that for xp or vista. But significant deviation is seen for land location. see attached figures. Another observation is elevation derived independently matches well for xp and vista runs while the ubuntu is under-estimating it. I have also attached the output dumps for both xp and ubuntu runs. I find lot of difference in meshing stage itself?! I am also attaching my test files in zipped format. ps: while xp and vista had identical python install files, ubuntu had "apt-get" installs. care is taken to clear cache to be sure all are clean runs. hope to catch you after Xmas. cheers -rajaraman On Thu, Dec 4, 2008 at 10:01 AM, wrote: > > Hi Rajaraman > > We have never come across such differences and we routinely use both > 32 and 64 bit platforms. > > Have you verified that compilation (compile_all.py) the unit test > suite (test_all.py) and the validation suite (validate_all.py) all > pass on both platforms? > > If they do and you still see the differences, we need to make some > more tests. > > Could I ask you to boil the script down to the simplest thing that > will show this difference and send them to us. > > We can then run them on our 32 bit and 64 bit platforms and verify the > outputs. > > > Cheers > Ole > > >> -----Original Message----- >> From: ramalingam rajaraman [mailto:rraman64@gmail.com] >> Sent: Thursday, 4 December 2008 4:05 >> To: Nielsen Ole >> Cc: S. Joseph Winston >> Subject: anuga in 64 bit vs 32 bit >> >> Dear Ole, >> >> We were testing anuga installed in two different system for >> reproducibility. Unfortunately one of them is 32 bit running under >> ubuntu with python2.5 and the other is 64 bit / suse / with python >> 2.4 (see the platform test results given below. >> >> both use anuga v5953 and the test run is with same set of files. I am >> attaching a sample time series. while peaks are more or less >> reproduced, the base elevation is not the same. >> >> have you come across such differences? >> >> can it be due to 64 bit vs 32 bit linux difference? >> >> cheers >> -rajaraman >> >> ------------------------------------------- >> 32bit box >> ------------ >> rraman@anuga:~$ python >> Python 2.5.2 (r252:60911, Jul 31 2008, 17:28:52) >> [GCC 4.2.3 (Ubuntu 4.2.3-2ubuntu7)] on linux2 >> Type "help", "copyright", "credits" or "license" for more >> information. >> >>> import platform >> >>> print platform.architecture() >> ('32bit', '') >> >>> print platform.machine() >> i686 >> >>> print platform.platform() >> Linux-2.6.24-21-generic-i686-with-debian-lenny-sid >> >>> print platform.processor() >> >> >>> print platform.python_version() >> 2.5.2 >> >>> print platform.version() >> #1 SMP Tue Oct 21 23:43:45 UTC 2008 >> >>> print platform.uname() >> ('Linux', 'anuga', '2.6.24-21-generic', '#1 SMP Tue Oct 21 23:43:45 >> UTC 2008', 'i686', '') >> >> -------------------------------------------- >> 64bit box >> ------------------- >> josephwinston:~ # python >> Python 2.4.5 (#1, Nov 13 2008, 15:06:52) >> [GCC 4.3.1 20080507 (prerelease) [gcc-4_3-branch revision 135036]] on >> linux2 Type "help", "copyright", "credits" or "license" for more >> information. >> >>> >> >>> print platform.architecture() >> ('64bit', 'ELF') >> >>> print platform.machine() >> x86_64 >> >>> print platform.platform() >> Linux-2.6.25.5-1.1-default-x86_64-with-SuSE-11.0-X86-64 >> >>> print platform.processor() >> x86_64 >> >>> print platform.python_version() >> 2.4.5 >> >>> print platform.version() >> #1 SMP 2008-06-07 01:55:22 +0200 >> >>> print platform.uname() >> ('Linux', 'josephwinston', '2.6.25.5-1.1-default', '#1 SMP 2008-06-07 >> 01:55:22 +0200', 'x86_64', 'x86_64') >> >>> >