HR: 09:15h
AN: U41A-04    [Abstracts]
TI: Computational Requirements For Large-Scale Seismic Modeling On PC-Based Clusters
AU: Blanch, J
EM: joakim@sensorwise.com
AF: SensorWise, Inc., 2908 Rodgerdale, Houston, TX 77042 United States
AU: * Cheng, A
EM: arthur@sensorwise.com
AF: SensorWise, Inc., 2908 Rodgerdale, Houston, TX 77042 United States
AU: Varsamis, G
EM: georgios@sensorwise.com
AF: SensorWise, Inc., 2908 Rodgerdale, Houston, TX 77042 United States
AB: With the development of deep reservoirs in the Gulf of Mexico, the proper modeling of seismic wave propagation in such an environment has become a huge computational challenge. A potential solution is the use of PC-based clusters. The theoretical peak speeds of these clusters are often impressive. However, the architectures of the clusters and the computers are seldom optimized for numerical computations. In particular the architecture can be very poor for the numerical solution of partial differential equations. Since the wave equation is at the core of many modeling and imaging algorithms, this may be a problem for many applications in exploration seismology. We have performed a series of tests to ascertain the attainable computational speed of PC clusters for large-scale Finite-Difference simulation of acoustic wave propagation. The results show that the most limiting factor is the internal memory bus, followed by the PCI bus. In general the actual peak speed is, by a factor of 4 or more, less than the theoretical peak speed for a single computer; and when the computation is performed in a cluster the actual peak is naturally less than for a single computer, and on the order of a factor of 10 or more less than the theoretical peak speed.
DE: 0902 Computational methods, seismic
DE: 3210 Modeling
DE: 3230 Numerical solutions
SC: Union [U]
MN: 2005 Joint Assembly