Petroleum is one of the main energy resources used in modern society and called the blood of the industry. Since petroleum resources are buried deep in the ground, seismic wave is commonly used in petroleum exploration. A seismic wave spreads differently in different strata and geologic structures and arrives at the detector at different times and from different directions with different delays. With these seismic data the conditions of oil and gas in the strata can be inferred inversely through computation. Usually hundreds of TB of seismic data are received in one exploration. The data received needs to be pre-processed, processed and interpreted, for which extremely high requirements for I/O throughput and floating point execution power are presented.
We capture run time features of seismic data processing by taking a certain migration processing module, as an example of one of the seismic data processing algorithms that takes the longest time to run.
Number of Shots
Seismic Data Volume
During seismic data processing, the CPU floating-point execution speed can reach about 400GFlops in a single node, inter-node network bandwidth of 700MB/s, storage data r/w speed of approximately 30~50MB/s, and local I/O throughput of approximately 10MB/s. Therefore, the seismic data migration processing application is floating point and network bandwidth intensive, and also requires relatively high I/O throughput.