PRESTACK MIGRATION SOLUTION
( Time & Depth 3D Kirchoff & Wave Equation Migration Software )
Our Prestack Migration Solution is a Beowulf PC cluster loaded with Prestack time and depth migration software. It utilizes a new computing environment we call Homogeneous Computing Architecture (HCA) which makes the cluster look and act like a single large mainframe computer. This results in a system with internal memory and distributed storage suitable for small to medium PSTM and PSDM jobs.
The system is designed to be installed into an office without additional power or air conditioning. . A raised-floor computer room is not required for our Prestack Migration System. HCA, therefore, is entirely scaleable. If you add another rack of cluster nodes, you also add enough additional seismic data storage to accommodate the extra compute power. The system is always balanced and efficient. The system software is what makes HCA possible.
A parallel file system distributes and stores the data efficiently on the disk drives within the cluster or external RAID. These are linked together to produce what appears to be a common file system to each node in the cluster. The PSTM and PSDM migration software has been written to take maximum advantage of the parallel file system. The system automatically moves seismic data to where it is needed without actually knowing where it is, in fact, stored. This happens in the background without operator or application program intervention. Since the input seismic data is "read only", it will be cached by the parallel file system, instantly ready just where it is needed.
To provide the operator with performance and activity information, the Prestack Migration Machine has a Beowulf monitor program with a graphical display. This shows the status of each node in the system and is continually updated with current information. Problems can be immediately found and corrective action taken.
PRESTACK MIGRATION SOFTWARE ( PSTM and PDSM )
Our Prestack Migration Solution comes equipped with Prestack Time Migration and Prestack Depth Migration seismic software. This is state-of-the-art migration software that has been optimized to run on a Beowulf cluster. It is based on the Kirchoff integral or isonochron method with all the modern trimmings. The PSDM module is driven by SPR raytracing for fast and accurate results. A full site of diagnostic and visualization software is also included. Both PSTM and PSDM modules include:
- Stretch Muting
- Double Aperture Control
- Point-To-Point Azimuthally-Varying Antialias Filtering
- Vsquared-T Amplitude Recovery
- Migration Fold Normalization
- Simplified 3D Geometry Input
Both programs input and output seismic data in SEGY disk file format, which interfaces to any of the common Seismic Processing systems. A full suite of SEGY tape and disk file utilities is also included.
CLUSTER CONSIDERATIONS For purposes of discussion, let's get the numbers for a typical 1 block deep-water offshore survey.
1 offshore block = (3*3) = 9 miles^^2
1 mile^^2 = (1609*1609) m^^2
input data is (12.5m * 40m) bins at 40 fold
=>1.9x10^^6 input traces/block
deep-water data (8 sec) = 8 kb/trace
=> 16 gb/block input data volume
Our target is to PSTM migrate a 100 block survey in 100 days; i.e., migrate at the rate of 1 block/day. Assume we have a cluster of 48 PC chassis with 2 nodes each or 96 nodes total.
PSTM 1 block/day = 20000 traces/node-day = 833 traces/node-hr
Therefore input data must move at a rate of:
833*8kb/3600 = 1.9 kb/sec into each chassis
The total input data volume (190x10^^6 traces) is 100 * 16gb = 1.6 tb which requires:
1.6 tb/72 gb = 22 disk drives (72 gb each)
1.6 tb/36 gb = 44 disk drives (36 gb each)
If we have two 36 gb drives/chassis (72 gb total), then the entire input data volume can easily be stored inside the cluster, in a parallel file system.
A mobile RAID system with 48 drives in it (maybe more for spares) would total 4 tb and be large enough to hold two 100 block surveys. With a "Gig E" connection, the RAID system could load the survey into the cluster in 1.6 tb/100 mb/sec = 5 hr. This is not a lot of latency to put up with in the event of a node storage failure.
The system requirements are:
48 dual CPU chassis with 1 gb memory and 3 (or 4) 32 gb disk drives
2 24-channel 100 bt e thernet switches
1 2 tb external mobile RAID system
This all assumes that PSTM will run at 833 input traces/node-hr which seems reasonable for 1200 mhz CPUs.
The output stack volume in 40m x 25m bins with 8 sec (8kb) traces will be:
1609 1609 ---- * ---- * 9 * 8kb = 186 mb / block
40 25 or 19 gb for the whole 3D survey. If this is kept in memory, it will require 1/2 gb / chassis. If output migrated gathers at every bin position are required as well then local disk storage will be necessary. Assuming 16 fold for the output gathers, this gives 6 gb per chassis. The total chassis storage requirements are:
memory: 1 gb for stack image + misc.
disk: 6 gb for gathers
disk: 72 gb for input data (if necessary)
Remember to multiply PSTM times by 2 in order to get PSDM estimates!!!