Reverse time migration (RTM) modeling is a critical component in the seismic processing workflow of oil and gas exploration. RTM imaging enables more accurate imaging in areas of complex structures and velocities by gathering a two-way acoustic image of seismic data in place of a one-way image. As this definition implies, it means creating and analyzing much more data with every seismic model, something that can be incredibly daunting — traditional simulations are already complex and require significant resources.
RTM comes in many flavors and is rapidly evolving. Its growth is driving increased geophysical processing requirements: vertical transverse isotropy, tilted transversely isotropic media, least squares/residual shot RTM (aka L2RTM, iteratively improving upon reflectivity) and elastic RTM, to name a few. And RTM can be seen as an important component or first step in more complicated (inversion) processes such as full-waveform inversion — where a process of iteration creates a high-resolution velocity using all of the information provided by the seismic waveform.
It is generally expected that a broad portfolio of RTM processing technologies will be a mainstay of seismic processing for many years to come — in fact, that RTM will be the high-end migration algorithm.
How will RTM be deployed?
Two trends will affect how RTM is deployed: an increase in algorithmic complexity and an increase in the volume, variety and fidelity of acquired data associated with seismic processing. These are called scale-up and scale-out, respectively.
Scale-up is the trend toward increased sophistication and functionality in RTM algorithms. An immediate consequence is that the compute power required for such algorithms increases too, irrespective of the data size. A subtler consequence is that algorithms such as L2RTM now tackle an optimization problem as well as a migration problem. Hence, scale-up also implies — if we envision our compute resources to be a cluster of some form — that there will be much more global data exchange between processes than is commonly done in today’s seismic processing schemes. Currently, much attention and effort are paid to keeping data movement to a minimum or to localize it as best as possible — whether it is I/O (within memory hierarchies and/or to storage) or message passing.
Scale-out is the almost exponential growth in seismic data acquisition. This growth is being driven by more complex acquisition geometries (wide azimuth, complete azimuth, coil acquisition), at better coverage, with longer recording times and at increased resolution — together with a desire to process seismic data at the resolution it is acquired.
Scale-out and scale-up feed on each other. The need for more accurate processing solutions (from acoustic to elastic, from migration to inversion) combined with the drive to higher resolution, specifically targeted and densely covered seismic data are forcing dramatic increases in processing, storage and networking requirements.
Impact on seismic processing organizations
Virtually every team in a seismic processing organization is affected by these simultaneous trends:
- Research – where geophysicists develop and maintain algorithms that anticipate compute technologies such as accelerators or coprocessors to provide much-needed processing power
- Development – where developers worry about feasibility and scalability/performance of the implementation, the programming environment and the integration of third party software and libraries.
- Systems – where IT staff — acutely aware of the realities of running day-to-day operations — migrate prototype development efforts to production workloads; providing support to the business units for a wide variety of processing services
- Facilities – where the CTO and IT asset owner are constantly evaluating how to maximize utilization, consolidate resources and provide remote processing capabilities and services; investigating how to best provide the most advanced but applied technologies as a service to the business units
How is Cray supporting RTM efforts?
At Cray, we have been involved in several efforts to address the intersection of seismic processing requirements and technology advances. Systems like the Cray® CS-Storm accelerator-optimized cluster system feature dense GPU platforms to deliver the acceleration needed for RTM simulations.