ATLAS' Fluctuations on Titan - ATLAS-Titan/misc GitHub Wiki
Atlas workload
The following plots depicts three experimental campaigns that have been computed in three different time intervals. Note that the data do not include any overhead from RP. All the instances of AthenaMP execute exactly the same simulation.
Experiment 1
The plot depicts the execution time of AthenaMP for simulation of the same 100 events (~77 minutes). AthenaMP instances have been sent for execution on TITAN by using Radical Pilot with different pilot sizes: 200, 500, 1000 and 2000 nodes. The number of CUs per pilot was equal to the number of nodes controlled by the pilot.

Experiment 2
The second plot depicts the execution time of AthenaMP for simulation of the same 16 events (~17 minutes). AthenaMP instances have been sent for execution on TITAN by using Radical Pilot with different pilot sizes: 256, 512, 1024 and 2048. The number of CUs per pilot was equal to five time the size of the pilot.

Experiment 3
The third plot depicts the execution time of AthenaMP for simulation of the same 16 events (~20 minutes). AthenaMP instances have been sent for execution on TITAN by using Radical Pilot with different pilot sizes: 256,512,1024 and 2048. The number of CUs per pilot was always equal to 2048.
