Changes between Version 3 and Version 4 of ticket/1821/BranchTesting

Timestamp:

2017-02-08T14:52:38+01:00 (8 years ago)

Author:

frrh

Comment:

--

ticket/1821/BranchTesting

@@ -112 +112 @@
 However we can't use this suite directly since it uses an old version of the MEDUSA branch which clashes with up to date versions of the GO6 branch! Are there any jobs around which use up to date versions of the GO6 AND MEDUSA branches because we can't test things without that? 
 
+Well we don't seem to have anything coupled-wise which fits the bill but Marc has u-aj369 which is an ocean-only run with MEDUSA using reasonably up to date versions of the MEDUSA and GO6 branches. So I take working copies of that - u-aj369control and u-aj369optalloc. These are then set up for 10 day runs with NEMO timer switched on and our optimisation branch applied in u-aj369optalloc. 
 
+Here's the results:
 
+The first thing to check is that both runs give the same results. tkdiff of solver.stat indicates bit comp at 10 days. 
+
+Control run:
+{{{ Total timing (sum) :
+ --------------------
+Elapsed Time (s)  CPU Time (s)
+        84009.892    72252.383
+ 
+Averaged timing on all processors :
+-----------------------------------
+Section             Elap. Time(s)  Elap. Time(%)  CPU Time(s)  CPU Time(%)  CPU/Elap Max elap(%)  Min elap(%)  Freq
+istate_init         0.2690187E+02   18.44            16.12      12.85        0.60         18.55     18.11       1.00
+tra_adv_muscl       0.1861153E+02   12.76            18.55      14.79        1.00         12.87     11.91     320.00
+trc_stp             0.8253362E+01    5.66             6.04       4.82        0.73          6.83      4.07     320.00
+sbc_ice_cice        0.8216911E+01    5.63             8.07       6.44        0.98          6.94      5.39     320.00
+tra_ldf_iso         0.7819895E+01    5.36             7.81       6.23        1.00          5.42      3.63     640.00
+trc_sms             0.7681783E+01    5.27             7.65       6.10        1.00          9.13      1.96     320.00
+sbc                 0.6609725E+01    4.53             4.82       3.84        0.73          8.93      3.17     320.00
+sol_pcg             0.5979100E+01    4.10             5.95       4.74        0.99          4.10      3.22     320.00
+trc_sbc             0.5684169E+01    3.90             5.65       4.50        0.99          7.42      0.41     320.00
+cice_sbc_init       0.3166722E+01    2.17             2.82       2.25        0.89          2.19      2.16       1.00
+trc_nxt             0.2791707E+01    1.91             2.78       2.21        1.00          2.11      0.97     320.00
+cice_sbc_out        0.2142690E+01    1.47             2.13       1.70        0.99          1.71      0.16     320.00
+dia_wri             0.1469471E+01    1.01             1.46       1.17        1.00          2.10      0.14     320.00
+ldf_slp             0.1370465E+01    0.94             1.36       1.09        0.99          0.97      0.92     320.00
+
+}}}
+
+Test Run
+
+{{{
+ Total timing (sum) :
+ --------------------
+Elapsed Time (s)  CPU Time (s)
+        67852.649    59359.526
+ 
+Averaged timing on all processors :
+-----------------------------------
+Section             Elap. Time(s)  Elap. Time(%)  CPU Time(s)  CPU Time(%)  CPU/Elap Max elap(%)  Min elap(%)  Freq
+tra_adv_muscl       0.1672964E+02   14.20            16.67      16.18        1.00         14.32     13.08     320.00
+sbc_ice_cice        0.8397770E+01    7.13             8.22       7.98        0.98          7.97      6.82     320.00
+trc_sms             0.7469511E+01    6.34             7.43       7.21        0.99         10.11      2.43     320.00
+trc_stp             0.6297795E+01    5.35             5.82       5.65        0.92          8.72      3.81     320.00
+trc_sbc             0.5423700E+01    4.60             5.37       5.21        0.99          9.02      0.28     320.00
+sbc                 0.5226975E+01    4.44             3.39       3.29        0.65          9.18      3.56     320.00
+sol_pcg             0.5086075E+01    4.32             5.06       4.91        0.99          4.32      3.23     320.00
+tra_ldf_iso         0.4719996E+01    4.01             4.71       4.57        1.00          4.09      1.83     640.00
+istate_init         0.3775559E+01    3.21             1.30       1.26        0.34          3.35      2.82       1.00
+trc_nxt             0.2831918E+01    2.40             2.82       2.73        0.99          2.59      1.40     320.00
+cice_sbc_init       0.2793284E+01    2.37             2.50       2.42        0.89          2.39      2.36       1.00
+cice_sbc_out        0.2308216E+01    1.96             2.29       2.23        0.99          2.26      1.10     320.00
+dom_vvl_rst         0.1516852E+01    1.29             0.35       0.34        0.23          1.45      0.83       3.00
+
+}}}
+
+So the overall elapsed time for our test run appears to drop by 20%. That's probably far enough outside the normal XC40 variability to suggest a positive impact. 
+A detailed look at the top routines above suggest wild variance in start up time (istate_init is the slowest routine in control but only about 20% of the cost in 
+the test run - that clearly is not a result of our optimisations!)
+
+tra_adv_muscl claims to be 20% speed up in elapsed time but slightly slower in CPU! Go figure. 
+sol_pcg claims to be slower in elapsed but faster in CPU!
+tra_ldf_iso claims ~25% speed up in elapsed and 35% speed up in CPU. 
+
+So as ever it's all pretty hazy but there's no evidence here to suggest things are any worse so it should be safe to go ahead with these changes..
+
+ 
+