Horizontal-vertical wrapper fine tuning
I created a copy of a34f
(a3di
) to see how the horizontal-vertical wrappers work and if it is responsive to changes in wallclock times. Everything worked more or less as expected, with some exceptions that I will explain.
This is a 10x10 case with a full wrapper:
This is a 100x100 case with 48-hour max wallclock so all job should fit in the wrapper. In fact, almost all the generations fit in, with the exception of the last four (this is difficult to see unless you enlarge the image). The only problem is that accounting for the times of the SIMs + POSTs (15min + 5min) there should be enough time in 48 hours to run the full thing inside one single job.
This is the same case after decreasing the max wallclock to 2-hour. Now you can see that there are more and smaller boxes.
Also in the GUI, the result is visible:
So Autosubmit takes into account changes in wallclock, but again my own calculation doesn't match. In 2 hour there should be enough time for 6 generations, and Autosubmit is only including 4 of them. I am running the experiment and this is what is happening, showing some output here:
8888 of 10101 jobs remaining (17:39)
Wrapper job a3dj_ASThread_160865513516_100_404 changed from SUBMITTED to status COMPLETED
Job a3dj_19600101_fc0012_1_SIM is COMPLETED
Job a3dj_19600101_fc0012_2_SIM is COMPLETED
Job a3dj_19600101_fc0012_3_SIM is COMPLETED
Job a3dj_19600101_fc0012_4_SIM is COMPLETED
Job a3dj_19600101_fc0012_5_SIM is COMPLETED
Job a3dj_19600101_fc0012_6_SIM is COMPLETED
......
......
......
Job a3dj_19600101_fc0015_20_SIM is COMPLETED
Job a3dj_19600101_fc0015_62_SIM is COMPLETED
Job a3dj_19600101_fc0015_POST is COMPLETED
8484 of 10101 jobs remaining (17:42)
Wrapper job a3dj_ASThread_1608655316347_100_404 changed from SUBMITTED to status COMPLETED
Job a3dj_19600101_fc0016_1_SIM is COMPLETED
Job a3dj_19600101_fc0016_2_SIM is COMPLETED
Job a3dj_19600101_fc0016_3_SIM is COMPLETED
Do you know @dbeltran @wuruchi if there is a margin/overhead taken into account? Could it be due to the member/chunk inverse order?