Writing on an Oracle blog, Melinda McDade provides some benchmark information from trials of Halliburton/Landmark’s ProMax seismic processing application running on a Sun cluster.
The test was run on with 48 X6275 server blades running in a Sun 6048 modular system connected by QDR Infiniband to a Lustre file system. The main tests involved scalability with increasing node count. For 3D prestack Kirchhoff time migration of a 71k trace dataset a 144x improvement was achieved going from 1 to 72 nodes. On a larger 283k trace volume, a 98x improvement was observed going from 1 to 96 nodes.
McDade attributes the ‘super linear’ scalability to hyperthreading (16 threads per node) and data caching. A 1.7x speedup was attributed to a recompilation of the source code for the current ProMax release using the latest Intel 11.1 compiler. McDade observes that current performance levels mean that ProMax can be run directly from Halliburton’s GeoProbe interpretation application to perform migrations on the fly, integrating logs and reservoir data from the OpenWorks database.
© Oil IT Journal - all rights reserved.