RiTMo, a Framework to Enhance RTM Productivity on HPC Resources

Simone Campagna (Cineca), Cristiano Calonaci (Cineca), Roberto Gori (Cineca), Nicola Bienati (ENI E&P), Clara Andreoletti (ENI E&P) 

73rd EAGE Conference & Exhibition, proceedings
75rd EAGE Conference & Exhibition, proceedings

Reverse Time Migration is a key algorithm for seismic imaging. On the other hand, RTM implies a huge consumption of computing resources (I/O, RAM, CPU) such that RTM runs have a deep impact on the overall exploitation of a production cluster. Moreover, a significant effort is required to administer thousands of single shot migrations and to collect partial results. Although the workflow is not awfully complicated, the huge amount of jobs and partial results needs particular attention, since any error resulting from hardware/software failures during one of the jobs can irreparably taint the final result. The main goal of the RiTMo framework is to simplify the execution of RTM projects, implementing the typical workflow in production environments where several other applications may be running concurrently. The RiTMo framework allows to easily configure and administer RTM projects; it is able to expose the full project as a single meta-job, hiding the underlieing complexity to the user. RiTMo is completely automatic, fault tolerant and reliable. Management of resources is flexible, as it can dynamically change the number of running jobs. Moreover, RiTMo enhances the overall throughput of the entire cluster, due to the efficient control of the resource usage. 

Link to download