SEP 2000 Abstract ======== The ideal software system for geoscience R&D should be functional for seismic and non seismic data analysis, processing and interpretation, yet open and flexible for unpredicted computing environment in user interface, graphics, and parallel computing. One way to do it is via a set of separate libraries with minimal dependency, and having the application program on top of the libraries free to choose some and not others. Who's on Top? ============= If the software environment is such that the application programmer writes the Unix main() and can optionally choose libraries to link to then the application is on top. If the application programmer writes one or some subroutines to be called by the system then the system is on top. The advantage of system on top is that it saves the application programmer the knowledge of what is going in the system. The disadvantage is that the system becomes more complex and less flexible when it comes to handling data flows that were not foreseen by the system designers. System programmers usually prefer the system on top, researchers and application programmers usually prefer the application on top. Seplib started in the late seventies with the application on top. When I started to program here in 1982, I found getpar in libc, where Rob Clayton put it. Getpar and some other utilities were libraries that the user might use if he wanted. In 1983 Stew Levin made a revolution and the system went on top; the application programmers started to #include and wrote a subroutine called MAIN() which was called by the main() in sep-main. It became very hard, practically impossible to use seplib without #including . It usually worked well, but if it did not and Stew was not around you were stuck. And it was a pain to debug because of the unusual #include and because so much was happening before your code has even started. (Other systems require application programmers to write two subroutines; one to be called for initial user interface and another to be called after everything is set up and when the data are flowing. This obviates the hclose() that was required in MAIN().) In 1984 Einar Kjartansson started SU while on a summer appointment at SEP. SU differed from SEPlib in two main features (1) it relied on trace headers while SEPlib relied on external header files (2) the application was on top. I liked SU better. I thought it was because I found the trace headers useful for processing field data with irregular geometry. Now I know that it was in spite of the trace headers which made SU specialized for seismic processing while SEPlib was more generic. It had nothing to do with the trace headers, it had everything to do with the fact that as an application programmer I prefered the application on top. Today the requirements from a software system for the geoscientist are much more extensive than they were in the early eighties. The system is required to be more flexible and handle other data types than 3-D cubes in which SEPlib was good at, and trace data in which SU was good at. Well log data, potential data, reservoir and velocity models, geographical information, and other examples, are neither cubes nor seismic traces. Flexibility is also required by the graphics and the graphical user interface in which X seems to be dominant (on Unix) but not supreme, its coverage of hard copy lacking and non standard, and there are competing widget sets and toolkits on top of X. Another reason for flexibility is that programming for parallel computing is not standard, and while many people are certain they know what the future is, it is certain that some of them are wrong. To keep our options open we should minimize the dependencies between components in the system. Two different applications may use the same command line interface (getpar) but one's data flow may be seplib cubes and the other's may be SEGY traces. A third application may input prestack data in SEGY traces and output a SEPlib cube stack. Two other application may share the same user and data interface but one would parallelize via MPI and the other via a Fortran 90 compiler. Schematically: /---------------\ /---------------\ /---------------\ | Application 1 | | Application 2 | | Application 3 | \---------------/ \---------------/ \---------------/ | \ \ / | \ / | | \ \----------\ | \---------\ / | | \ / \ | \ / | | \ / \ | \ / | /-----------\ /-----------\ /-----------\ /-----------\ /-----------\ | Library 1 | | Library 2 | | Library 3 | | Library 4 | | Library 5 | \-----------/ \-----------/ \-----------/ \-----------/ \-----------/ A system which is a big blob of inter-dependent functions which is an all or nothing choice is less flexible and the choice might very well be nothing because one component does not fit. This also has the advantage of easier distributed development. Components ========== /------------- Command Line User Interface ------------------------------\ | | | Consolidate Getpar, Getoption, Self-doc, Error/Warning diagnostics | CLUI | | \------------------------------------------------------------------------/ /------------- Graphical User Interface ---------------------------------\ | | | Direct for interactive programs | | | | Job builder for batch programs | GUI | | | Tcl/Xtpanel/CGI/Java built-ins/ ... | | | \------------------------------------------------------------------------/ /------------- Graphics -------------------------------------------------\ | | | Vplot/Widget set/... | | | \------------------------------------------------------------------------/ /------------- Math -----------------------------------------------------\ | | | BLAS, FFT, convolutions, Wiener-Levinson, Interpolation, ... | | | \------------------------------------------------------------------------/ /------------- Solvers and optimizers -----------------------------------\ | | | Vectors (N-D cubes, scatters, mixed axis, ...) with I/O methods! | | | | Operators (Matrices, Sparse matrices, Aprod) optional I/O methods | | | HCL? | Solvers (Page Saunders, cgstep, IRLS, ...) | | | | Optimizers (Cyclic Coordinates, Gauss Seidell, CG, Monte-Carlo, ...) | | | \------------------------------------------------------------------------/ /------------- SEGY (SU) ------------------------------------------------\ | | | Gettr, Puttr, SEGY Header utilities, applications, GUI job builder | SU++ | | \------------------------------------------------------------------------/ /------------- LIS/LAS/DLIS ---------------------------------------------\ | | | I/O to/from scatters | | | \------------------------------------------------------------------------/ /------------- Parametric Modeling --------------------------------------\ | | | XoX or alike | | | \------------------------------------------------------------------------/