In the oil and gas industry, subsalt structure imaging has been studied for years. True 3D prestack depth migration has proved to be the most useful method in subsalt imaging, but it is also the most costly method (Sheriff an d Geldart, 1995). Since 3D data sizes are very large, it takes months to do 3D prestack depth migration on a conventional computer. 3D prestack depth migration jobs have rarely been done in small sized data processing companies because of the high price o f a supercomputer. A Beowulf-like supercomputer (Becker et. al., 1995) has been constructed by using pile-of-PCs. This allows one to do 3D prestack depth migration at low cost and with high quality. This thesis describes the tape reading phase in the 3D p restack depth migration software package. The tape reading process has been a basic issue in exploration seismic data processing for many years. Because the seismic data volume is large, heavy I/O is a bottleneck for data processing. The basic design and implementation of a successful tape readin g process are given in this thesis. The tesing data have been preprocessed in our Beowulf-like supercomputer environment. The timings for different preprocesses are determined. One can see that threading saves about 40 percent of the real elapsed time, muting, resampling, and packing save more time, while filtering and byte swapping cause the process to take more time. A customized trace, with a cust omized header in each trace, has been generated before sending the trace to the migration nodes. A bisic migration unit, GSP, which is grouped with data in nearby shot points, is generated by the tape reader.