To provide the interactive imaging capabilities discussed above, the imaging hardware must provide certain minimum levels of performance. Figure 18.5 schematically represents data flowing from the trace manager to the computational processes and the image generated there being sent to the display manager for eventual display on the monitor. To perform the interactive focusing discussed above, the computational engine must deliver approximately 200 to . While this number may be difficult and expensive to obtain in a single-processor system, it is easily obtainable in parallel systems. Likewise, in order to satisfy the demands of interactive stacking, the trace manager is required to deliver many thousands of traces per second (approximately 4 to 8 Kbytes/trace) to the computational processes. Since these traces are essentially randomly ordered throughout a multigigabyte data volume, a simple calculation will show that, for current disk drive technology, the limiting factor in supplying the data is the disk seek time, not the aggregate transfer rate. Again, a solution to the problem is for a number of disks working in parallel to provide the needed performance. Finally, in order to display movies of seismic gathers at eight frames per second, the graphics processors must be able to absorb and display eight megabytes of data per second. Once again, this requirement may be satisfied by multiple nodes working in parallel.
In addition to the general performance issues, which could be achieved by the creation of a specially built machine, or the addition of custom I/O devices to an existing supercomputer, we chose to use only commercially available hardware. The reasons for this choice are twofold: We wanted other interested researchers to be able to duplicate our efforts, and we wanted the system to be reasonably affordable.