I ran benchmarks on laptop (elnino), desktop (soot), supercomputer
(esmf) for the first time in some weeks.
File creation and benchmark times seem much slower that I expected.
File creation is sickeningly slow on laptop and desktop,
takes thousands of seconds. Faster on ESMF. Same on
benchmarks: very slow on laptop and desktop, faster on ESMF
interactive node.
Sorry this is so late - I was going to andwer last night and then ended up doing a lot of debugging on what I thought the answer was. I found partial answers (the ncap that was dong the file creation was not directed specifically at the most recent build, but would have picked up the system ncap preferentially - if the system ncap was pre-henry-fix, it would be quite slow.
However, the file build times that I got on bonk for the most recent nco is reasonably fast. Here's the result of it running on bonk (2x700MHz-PIII).
NCO Test Result Summary for:
Linux www.tacgi.com 2.6.7 #2 SMP Mon Jul 26 11:00:35 PDT 2004 i686 GNU/Linux
using: gcc (GCC) 3.4.4 20050314 (prerelease) (Debian 3.4.3-12)
Test Total Wallclock Time (s)
=====================================================
Creating gne_exp: 0.8825
Populating gne_exp: 9.7181
Creating sml_stl: 3.2579
Populating sml_stl: 20.5723
Creating stl_5km: 8.1514
Populating stl_5km: 64.3846
Creating ipcc_dly_T85: 36.4252
Populating ipcc_dly_T85: 1159.3322
ando n sand, it seems to be trucking along pretty well.
NCO Test Result Summary for:
Linux sand.ess.uci.edu 2.6.10-5-amd64-generic #1 Tue Jun 7 09:14:47 UTC 2005 x86_64 GNU/Linux
using: gcc (GCC) 3.4.4 20050209 (prerelease) (Debian 3.4.3-9ubuntu4)
Test Total Wallclock Time (s)
=====================================================
Creating gne_exp: 0.2491
Populating gne_exp: 1.7909
Creating sml_stl: 0.4859
Populating sml_stl: 6.1456
Creating stl_5km: 1.8815
Populating stl_5km: 13.7616
Creating ipcc_dly_T85: 6.1816
Populating ipcc_dly_T85: 441.1572
NCO Test Result Summary for:
AIX esmf04m 1 5 0020D3FA4C00 unknown unknown AIX
using: xlc version ??
Test Total Wallclock Time (s)
=====================================================
Creating gne_exp: 0.3734
Populating gne_exp: 5.6133
Creating sml_stl: 0.9584
Populating sml_stl: 12.9974
Creating stl_5km: 2.9033
Populating stl_5km: 40.7807
Creating ipcc_dly_T85: 23.4310
Populating ipcc_dly_T85: 182.9174
The above problem with ncap might be the correct answer for the file creation tests. I'm still debugging the rest of the benchmarks.
I'm sending you the latest version of the md5 checksum script (variant of nco_bm.pl).. It's different enough that I didn't want to commit it until I was pretty dang sure it was working correctly so it wouldn't screw up all the platforms.
However, while it runs on all platforms, I'm getting some strange effects with ncap on sand (while it's up) in the regression tests and on the esmf - some of them seem to be overlapping. More on this in a separate thread.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I ran benchmarks on laptop (elnino), desktop (soot), supercomputer
(esmf) for the first time in some weeks.
File creation and benchmark times seem much slower that I expected.
File creation is sickeningly slow on laptop and desktop,
takes thousands of seconds. Faster on ESMF. Same on
benchmarks: very slow on laptop and desktop, faster on ESMF
interactive node.
Anybody know what's going on?
These benchmarks are all with NCO 3.0.1:
ESMF no args:
ncap: 1 1 135.6445
ncbo: 0 1 1 0.0597
ncea: 1 1 21.7713
ncpdq: 2 2 116.8330
ncra: 1 1 111.6358
ncwa: 3 3 211.5128
ESMF thr_nbr=1
Test Success Failure Total Time
ncap: 1 1 176.7508
ncbo: 1 1 24.1652
ncea: 1 1 34.6227
ncpdq: 2 2 164.6700
ncra: 1 1 125.9528
ncwa: 3 3 255.2852
ESMF thr_nbr=2
Test Success Failure Total Time (OMP threads = 2)
ncap: 1 1 202.9640
ncbo: 1 1 60.2211
ncea: 1 1 36.2658
ncpdq: 2 2 166.1194
ncra: 1 1 144.9846
ncwa: 3 3 266.9105
ESMF thr_nbr=4
Test Success Failure Total Time (OMP threads = 4)
ncap: 1 1 138.8941
ncbo: 1 1 34.9203
ncea: 1 1 21.2575
ncpdq: 2 2 116.1546
ncra: 1 1 105.6453
ncwa: 3 3 212.8289
soot thr_nbr=1
Test Success Failure Total Time
ncap: 1 1 1011.9080
ncbo: 0 1 1 0.6274
ncea: 1 1 119.5755
ncpdq: 2 2 783.5053
ncra: 1 1 718.2057
ncwa: 3 3 903.4869
soot thr_nbr=2
Test Success Failure Total Time (OMP threads = 2)
ncap: 1 1 1005.6657
ncbo: 1 1 801.2006
ncea: 1 1 127.3837
ncpdq: 2 2 1087.1874
ncra: 1 1 1042.5778
ncwa: 3 3 609.0312
elnino no args:
Test Success Failure Total Time
ncap: 1 1 350.1355
ncbo: 1 1 135.5039
ncea: 1 1 46.4986
ncpdq: 2 2 303.5463
ncra: 1 1 2345.6275
ncwa: 3 3 2630.3316
Hi Charlie,
Sorry this is so late - I was going to andwer last night and then ended up doing a lot of debugging on what I thought the answer was. I found partial answers (the ncap that was dong the file creation was not directed specifically at the most recent build, but would have picked up the system ncap preferentially - if the system ncap was pre-henry-fix, it would be quite slow.
However, the file build times that I got on bonk for the most recent nco is reasonably fast. Here's the result of it running on bonk (2x700MHz-PIII).
NCO Test Result Summary for:
Linux www.tacgi.com 2.6.7 #2 SMP Mon Jul 26 11:00:35 PDT 2004 i686 GNU/Linux
using: gcc (GCC) 3.4.4 20050314 (prerelease) (Debian 3.4.3-12)
Test Total Wallclock Time (s)
=====================================================
Creating gne_exp: 0.8825
Populating gne_exp: 9.7181
Creating sml_stl: 3.2579
Populating sml_stl: 20.5723
Creating stl_5km: 8.1514
Populating stl_5km: 64.3846
Creating ipcc_dly_T85: 36.4252
Populating ipcc_dly_T85: 1159.3322
ando n sand, it seems to be trucking along pretty well.
NCO Test Result Summary for:
Linux sand.ess.uci.edu 2.6.10-5-amd64-generic #1 Tue Jun 7 09:14:47 UTC 2005 x86_64 GNU/Linux
using: gcc (GCC) 3.4.4 20050209 (prerelease) (Debian 3.4.3-9ubuntu4)
Test Total Wallclock Time (s)
=====================================================
Creating gne_exp: 0.2491
Populating gne_exp: 1.7909
Creating sml_stl: 0.4859
Populating sml_stl: 6.1456
Creating stl_5km: 1.8815
Populating stl_5km: 13.7616
Creating ipcc_dly_T85: 6.1816
Populating ipcc_dly_T85: 441.1572
NCO Test Result Summary for:
AIX esmf04m 1 5 0020D3FA4C00 unknown unknown AIX
using: xlc version ??
Test Total Wallclock Time (s)
=====================================================
Creating gne_exp: 0.3734
Populating gne_exp: 5.6133
Creating sml_stl: 0.9584
Populating sml_stl: 12.9974
Creating stl_5km: 2.9033
Populating stl_5km: 40.7807
Creating ipcc_dly_T85: 23.4310
Populating ipcc_dly_T85: 182.9174
The above problem with ncap might be the correct answer for the file creation tests. I'm still debugging the rest of the benchmarks.
I'm sending you the latest version of the md5 checksum script (variant of nco_bm.pl).. It's different enough that I didn't want to commit it until I was pretty dang sure it was working correctly so it wouldn't screw up all the platforms.
However, while it runs on all platforms, I'm getting some strange effects with ncap on sand (while it's up) in the regression tests and on the esmf - some of them seem to be overlapping. More on this in a separate thread.