AWS

SeisWave has successfully partnered with AWS to migrate its Reverse Time Migration (RTM) technology into cloud computing. By leveraging the capabilities of AWS cloud computing, SeisWave RTM has achieved high efficiency and robust fault tolerance. Two projects have been completed under this partnership: one involving a real ocean bottom node (OBN) product, and the other using the SEAM model. Both projects have achieved high-quality migration results at remarkably low costs.

2023 Rich HPC --- a real OBN project

In November 2022, SeisWave received a dataset from the Gulf of Mexico, consisting of actual production data totaling 35 terabytes. The migration velocity model used was a Tilted Transversely Isotropic (TTI) model. SeisWave completed the entire project within one month, from data preparation to the final project report. The imaging output was presented at the 2023 Rich HPC Conference, garnering widespread interest due to its high quality and cost-effectiveness."

Project Milestone Timeline

02/10：Received Data (~1700 shots, ~35TB, TTI Model)
Single node benchmark
02/14: data conversion and QC.
02/15: migration and QC.
02/16-02/19: Full survey QC; 428 node selection, pre-processing and QC.
02/20-02/22: AWS environment setting up and benchmark credit application.
02/23-02/25: 40Hz RTM preparation and execution.
02/26: Project report preparation
02/28: Got showrights approval
03/01: Presented the results at Rice HPC conference

The RTM job was completed in around 15 hours using up to 200 machines (machine number was automatically adjusted according to the requirements of the job and the availability of resources). As indicated in the table below, the Graviton 2 machine was chosen for its excellent cost-performance ratio. All the machines employed were spot instances, which significantly reducing our costs. The total computing cost amounted to only about $1,100. The following sections will detail the main RTM parameters, the machine selection process, and the final performance outcomes. Please note that the final images will not be shown here due to show rights.

RTM key parameters

TTI RTM mirror imaging
Input:
- 428 nodes
- 8km maximum offset
- 10.5s data
- TTI velocity model

Migration:
- Mirror imaging
- 10km aperture
- 40hz

Output:
- Depth: 13km
- DZ: 8m
- CDP: 15m×15km

Project Performance:

Computing resource: c7g.8xlarge
Run time: ~15hrs/node
Price: $2.503/node
Total price: $1100

2023 AWS Symposium --- SEAM project

Through our partnership with AWS and Bluware, we successfully completed an entire velocity update cycle in just 10 hours - a process that typically takes several weeks.

interview on AWS Energy Symposium

Poster on AWS symposium

2024 Rice HPC

To evaluate the Reverse Time Migration (RTM) capability for high-frequency migration, we conducted an 80 Hz migration test on a 3D Tilted Transverse Isotropy (TTI) model. The results displayed significantly more detailed information compared to the 40 Hz RTM results. Although the cost was approximately ten times higher, the enhanced detail justifies the increased expenditure.

In partnership with AWS, we conducted RTM migration tests on various types of hardware using production 3D data to evaluate their performance. To ensure a comprehensive comparison, we selected two AMD machines, two Intel machines, and two Graviton machines. The efficiency of these machines was measured by analyzing both the running time and the cost per shot. During these tests, the Graviton machines, especially Graviton 3, demonstrated a significant advantage, especially for lower vcpus.