In petroleum mining, drilling expenses are very high. Drilling an oil well costs hundreds of millions of Yuan. The petroleum company will suffer tremendous losses if the petroleum resources are acquired only through drilling during the early-stage exploration. The common practice today involves the use of complex earthquake imaging methods to create a detailed 3D map of hidden oil rocks and natural gas reserves before drilling so as to increase the success rate of drilling and hence reduce the exploration risk. The internationally-recognized advanced technology today is the offset imaging technology that provides accurate structural imaging. This technology has been extensively applied in the petroleum exploration and development sectors.
3D pre-iteration offset technology involves massive data load and computing scale. Therefore, the computing cluster system needs to have a robust parallel computing and processing capability and data storage capability, with a computing scale of at least 24 coverages and 2,400,000 3D seismic data and 3D pre-overlap depth offset in 100sqkm and a total I/O of at least 2.5T. The computing and processing time of such a scale needs to be within 10 hours.
Analysis of Application Features
The Institute’s software system comprises numerous self-compiled software for manual seismic data processing of depth offset system and visualized treatment of 3D seismic data other than the OMEGA system of Western Geophysical and ADSView of Hayden. The GPU cluster under this procurement project is expected to facilitate the pace of scientific research and the process of industrialization of scientific research achievements via its robust processing capacity.
In accordance with the features of petroleum exploration and application development, the Institute established very high requirements on the five aspects of parallel computing capacity and cluster management capacity, and specifically included robust parallel computing and processing capacity; easily operable and manageable cluster management system with a graphical interface; open infrastructure and application platform; stringent quality assurance and good service ability and service system; rapid response to and timely settlement of system faults; convenient maintenance and management of the whole system for smooth application and transition of new technologies and products in the system and for convenient upgrading of the system. To summarize, the system requirements entail high availability, ease of use and convenient management on the one hand and scalability on the other. First, an open architecture and application platform is needed for future upgrading and regeneration. Second, the system needs to conveniently maintained and managed for the ease of system upgrading.
The Institute adopts Inspur Tiansuo TS10000 high-performance cluster server system which is furnished with advanced high-speed interconnection technology and an open modular system to satisfy the Institute’s requirements on parallel computing and processing capacity and cluster management capacity of the system.
The core architecture of Inspur Tiansuo TS10000 entails the high-speed computing module, high-speed exchange module, storage module and management module (see the following network topology).
1. Computing module
The cluster system comprises 24 GPU nodes and the peak speed of CPU floating-point computation reaches 5TFLOPS theoretically (floating-point computation speed per second); 96 NVIDIA GTX680 acceleration cards are also furnished and the computing performance is substantially improved to satisfy the needs for large-scale parallel computation.
2. Management storage module
The management storage module is liable for system hardware resources monitoring, as well as system resources and data storage of the whole system. It features robust service processing ability. The cluster management software provides a graphic user management interface, high-performance communication library and integrated software tool.
The “3D seismic pre-iteration depth offset processing” and “3D image processing” hardware system as designed by Inspur, China's leading server supplier, and successfully applied by the Sinopec Research Institute of Petroleum Exploration fully demonstrates the advantages of the high-performance computing cluster with an open architecture. It also provides a mature model for many scientific research sectors including that of petroleum exploration.
The Inspur Tiansuo TS10000 system designed for the institute complies with the design concept of “flexible deployment”. It features a computing speed of 1 trillion and offers unique advantages in manageability and scalability.