The CRAY X-MP/416 was used to multitask 3-D forward modeling by the high-order finite difference method. Flowtrace analysis reveals that the most expensive operation in the unitasked program is a matrix vector multiplication. The in-core and out-of-core versions of a reentrant subroutine can perform any fraction of the matrix vector multiplication independently, a pattern compatible with multitasking. The matrix vector multipli cation routine can be distributed over two to four pro cessors. The rest of the program utilizes the micro tasking feature that lets the system treat independent iterations of DO-loops as subtasks to be performed by any available processor. The availability of the Solid- State Storage Device (SSD) meant the I/O wait time was virtually zero. A performance study determined a theo retical speedup, taking into account the multitasking overhead. Multitasking programs utilizing both macro- tasking and microtasking features obtained actual speedups that were approximately 80% of the ideal speedup.
[1]
Ernst L. Leiss,et al.
An Experimental Implementation of Migration Algorithms On the Intel Hypercube
,
1987
.
[2]
R. M. Alford,et al.
ACCURACY OF FINITE‐DIFFERENCE MODELING OF THE ACOUSTIC WAVE EQUATION
,
1974
.
[3]
Kai Hwang,et al.
Computer architecture and parallel processing
,
1984,
McGraw-Hill Series in computer organization and architecture.
[4]
Jon F. Claerbout,et al.
Coarse grid calculations of waves in inhomogeneous media with application to delineation of complicated seismic structure
,
1970
.
[5]
Jean-Loup Baer,et al.
Computer systems architecture
,
1980
.
[6]
Philip F. Ridler,et al.
Fortran Reference Manual
,
1979
.
[7]
O. Johnson,et al.
High order finite-difference modeling and reverse time migration
,
1985
.