PARALLEL COMPUTING

Reflection seismology in general, and SEP in particular, is always dealing with larger datasets and implementing more more complex algorithms to image these datasets. SEP has a long tradition of parallel computing to tackle these problems. Currently SEP has several options available:

• CM5 (vesuvio): Uses message passing and CM Fortran. The continuing status of the CM at SEP is in doubt, and the processor (25 Mhz for each of the 32 processors) makes learning to code on it a questionable decision.
• Cray: We have access to Cray supercomputers as part of our sponsorship agreement. In general this is an underutilized option.
• SGI Power Challenge (santorin): This is the most convenient path to parallel coding. The Power Challenge uses a shared memory model that enables you to code essentially the same as on other machines. The Power Challenge provides automatic parallelization, which means you don't have to do any work to start taking advantage of multiple processors. This automatic parallelization has its limits. It is not allowed in C++, limited in C, and sometimes deceiving in Fortran 77 and Fortran 90. You can also use explicit directives (check the on-line documentation /usr/sbin/insight for more information).