GPU SCHEDULER EXPERIMENTS Following testflow was used to produce the results discussed in Section 3.5.3.2 of the AI-SPRINT deliverable D3.5 The Gpu scheduler has been started using andreasOnIM.radl file via Infrastracture Manager (IM and AWS accounts are needed) When the platform was already running the first step was to upload necessary files needed for training. Input data and jobs scripts were uploaded using script: ./init_upload.sh Then the profiling was launched using scripts (next script was lauched after the precedent one finished all the jobs): ./init_M60_part_1.sh ./init_M60_part_2.sh ./init_T4.sh The real training consisted of launching script: ./real_training.sh the output of the jobsManager during training part can be found in the file: jobsManager.log the output of the real_training.sh can be found inside testingScriptOutput.txt Note that to use the scripts you need to change inside the ip of the control unit (VM containg jobs manager)