Main.HPCxTestRuns History
Hide minor edits - Show changes to output
Changed lines 10-11 from:
Equivalent runs on [=HPCx=] and Bristol quest cluster compared using @@tarograd.pro@@.
to:
Equivalent runs on [=HPCx=] and quest cluster compared using @@tarograd.pro@@.
Changed lines 35-36 from:
to:
Changed lines 8-14 from:
!!xbyvs
We compared equivalent 100 year runs on [=HPCx=] and ormen at Bristol, starting from identical restarts. The results were compared using the IDL program @@tarograd.pro@@ (available on request).
Output:
to:
!!xcggb
Equivalent runs on [=HPCx=] and Bristol quest cluster compared using @@tarograd.pro@@.
Statistics:
Deleted lines 14-15:
Ormen (xciib) v HPCx (xbdho)
Changed lines 16-22 from:
temperature at 1.5m 1.023
total precipitation rate kg/m2/s 0.724
pressure at mean sea level -1.256
potential temperature (ocean) deg.c 1.421
salinity (ocean) (psu-35)/1000 1.202
aice : ice concentration -0.882
to:
temperature at 1.5m 0.468
total precipitation rate kg/m2/s 0.014
pressure at mean sea level -0.545
potential temperature (ocean) deg.c -0.024
salinity (ocean) (psu-35)/1000 0.637
aice : ice concentration -0.774
Changed lines 24-30 from:
temperature at 1.5m 0.076
total precipitation rate kg/m2/s 0.068
pressure at mean sea level 0.043
potential temperature (ocean) deg.c 0.078
salinity (ocean) (psu-35)/1000 0.056
aice : ice concentration 0.084
to:
temperature at 1.5m 0.055
total precipitation rate kg/m2/s 0.062
pressure at mean sea level 0.068
potential temperature (ocean) deg.c 0.055
salinity (ocean) (psu-35)/1000 0.084
aice : ice concentration 0.112
Added lines 32-63:
Plots:
* [[http://www.famous.ac.uk/Stats/tarograd_xcggb_xcggc.ps|download]]
!!xbyvs
We compared equivalent 100 year runs on [=HPCx=] and ormen at Bristol, starting from identical restarts. The results were compared using the IDL program @@tarograd.pro@@ (available on request).
Statistics:
[@
Ormen (xciib) v HPCx (xbdho)
t-statistic for global average:
temperature at 1.5m 1.023
total precipitation rate kg/m2/s 0.724
pressure at mean sea level -1.256
potential temperature (ocean) deg.c 1.421
salinity (ocean) (psu-35)/1000 1.202
aice : ice concentration -0.882
Proportion of area with |t|>2
temperature at 1.5m 0.076
total precipitation rate kg/m2/s 0.068
pressure at mean sea level 0.043
potential temperature (ocean) deg.c 0.078
salinity (ocean) (psu-35)/1000 0.056
aice : ice concentration 0.084
@]
Changed lines 11-12 from:
We compared equivalent 100 year runs on HPCx and ormen at Bristol, starting from identical restarts. The results were compared using the IDL program @@tarograd.pro@@ (available on request).
to:
We compared equivalent 100 year runs on [=HPCx=] and ormen at Bristol, starting from identical restarts. The results were compared using the IDL program @@tarograd.pro@@ (available on request).
Changed lines 1-5 from:
! Statistical Comparisons
Statistical comparsions for each FAMOUS release
to:
! Test Runs & Statistical Comparisons
Comparisons of equivalent FAMOUS runs on different platforms.
Added line 3:
Added line 10:
Changed lines 41-42 from:
to:
Changed lines 1-17 from:
!!Test Runs
There are results for five 100-year runs on [=HPCx=].
# '''xbyjc''' - This is the "base" FAMOUS job on [=HPCx=]. It was set up to be as close as possible to Simon's "base" Ruby job '''xbxxy''', which is in turn based on the Met Office's adtan.
# '''xbyje''' - As the base job but with reconfiguration of the atmosphere and ocean start files.
# '''xbyjh''' - As the base job but with the optimisation level reduced from O3 to O2.
# '''xbyji''' - This job starts from atmosphere and ocean restart dumps produced from a 1 year run. The date of the restart file is put back one year to reconcile it with the model start date. This is done using the reconfiguration, which also alters some of the data fields.
# '''xbyjn''' - As above, the job starts from 1 year restart dump files, however in this case the dates are altered using the @@change_dump_data@@ utility which changes the date only.
!!Ensemble Results
Statistics produced by Jonathan Gregory's program @@compare_climates.pro@@, which does a statistical test using the variability within the ensemble to test whether the ensemble mean differs from the reference run (adtan). This is better than using temporal variability as a substitute for intraensemble variability because it avoids have to make assumptions about lack of correlation.
to:
! Statistical Comparisons
Statistical comparsions for each FAMOUS release.
----
!!xbyvs
We compared equivalent 100 year runs on HPCx and ormen at Bristol, starting from identical restarts. The results were compared using the IDL program @@tarograd.pro@@ (available on request).
Output:
Added lines 14-55:
Ormen (xciib) v HPCx (xbdho)
t-statistic for global average:
temperature at 1.5m 1.023
total precipitation rate kg/m2/s 0.724
pressure at mean sea level -1.256
potential temperature (ocean) deg.c 1.421
salinity (ocean) (psu-35)/1000 1.202
aice : ice concentration -0.882
Proportion of area with |t|>2
temperature at 1.5m 0.076
total precipitation rate kg/m2/s 0.068
pressure at mean sea level 0.043
potential temperature (ocean) deg.c 0.078
salinity (ocean) (psu-35)/1000 0.056
aice : ice concentration 0.084
@]
Plots:
* [[http://www.famous.ac.uk/Stats/tarograd_xcggb_xcggc.ps|download]]
----
!!adtan Test Runs
There are results for five 100-year runs on [=HPCx=].
# '''xbyjc''' - This is the "base" FAMOUS job on [=HPCx=]. It was set up to be as close as possible to Simon's "base" Ruby job '''xbxxy''', which is in turn based on the Met Office's adtan.
# '''xbyje''' - As the base job but with reconfiguration of the atmosphere and ocean start files.
# '''xbyjh''' - As the base job but with the optimisation level reduced from O3 to O2.
# '''xbyji''' - This job starts from atmosphere and ocean restart dumps produced from a 1 year run. The date of the restart file is put back one year to reconcile it with the model start date. This is done using the reconfiguration, which also alters some of the data fields.
# '''xbyjn''' - As above, the job starts from 1 year restart dump files, however in this case the dates are altered using the @@change_dump_data@@ utility which changes the date only.
!!!Ensemble Results
Statistics produced by Jonathan Gregory's program @@compare_climates.pro@@, which does a statistical test using the variability within the ensemble to test whether the ensemble mean differs from the reference run (adtan). This is better than using temporal variability as a substitute for intraensemble variability because it avoids have to make assumptions about lack of correlation.
[@
Changed lines 88-92 from:
!!Individual Comparisons
The decadal means for each test run were compared against adtan results using adtan - "compares global average timeseries and fields of some basic quantities from corresponding meaning periods of two sources".
to:
!!!Individual Comparisons
The decadal means for each test run were compared against adtan results using tarograd.pro - "compares global average timeseries and fields of some basic quantities from corresponding meaning periods of two sources".