Files
4_version_offload_nvme
Folders and files
Name | Name | Last commit date | ||
---|---|---|---|---|
parent directory.. | ||||
COMPILATION WORKFLOW: - Source (or create if it is not available for your system) the environment file. Ej. "source environment_mn.sh" for the MareNostrum case. - Make. This should create 3 binaries. EXECUTION WORKFLOW: 1*Edit these two configuration files: - fwi_frequencies.txt is a list of wavelef frequencies. One freq. value per line. Be careful, the number of cells in the domain is multiplied by 8 everytime the frequency is incremented by 1 Hz. - fwi_params.txt: simulation definition parameters. - z dimension in meters - x dimension in meters - y dimension in meters (in real world, this is the largest) - vmin: min velocity propogation (it does not make sense to change this) - srclen: unused - rcvlen: unused - nshots: number of shots to be processed. Each shot will be managed by one slave, so, in practice this is the number of slave nodes. - ngradients: number of gradient iterations. Between 2 and 25 in real cases. - ntests: number of test iterations: Between 2 and 6 in real cases. - slavemem: slave's node memory in GB - workermem: worker's memory in GB ** Notice that in real world FWI the number of shots will be in the order of thousands to tens of thousands. ** The domain is divided according to the worker's memory, less memory equals more workers. 2*Run the schedule. This will generate a ../SetupParams/fwi_schedule.txt 3*The last column of the ../SetupParams/fwi_schedule.txt file is the number of workers needed for each case. The number of shots to process is the number of slave nodes. Edit the job file accordingly. 4*Submit the job. Have fun :)