No label defined (Q4026764)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 125114
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 125114 |
Statements
21 February 1993
0 references
scientific computation
0 references
reconfigurable programs
0 references
0.7266137003898621
0 references
0.7153110504150391
0 references
0.7138307094573975
0 references
0.7130237221717834
0 references
[For the entire collection see Zbl 0746.00068.] Folklore from the early days of our field has it that problem solvers, who employed numerical approximation techniques, could regularly halt their computer in order to manually inspect the progress of their program. Based upon what they saw, and knowing the layout of their program in memory, the user could even alter a few parameter -- say, step sizes or tolerances -- in order to adapt the algorithm's behavior. Since those early days, we have evolved higher-level programming languages, more powerful computers, and better user interface software than ever before. But the level of dynamic control that the user has over his program is still at the level of halting, selective `tweaking' of values, and resumption of execution. Only small-scale adaptations can be performed, since current software technology requires the program structure to remain essentially the same. But what if -- dynamically -- the user discovers that an alternate method altogether could take over computation and perform better? Even though the intermediate program state might be easily used by an alternate program component (e.g., integrators often use the same data structures but vary only in the order they update grid elements), the user cannot change methods unless he has explicitly designed this capability into his program right from the start. Anticipating all possible alternatives that might arise in a long-running scientific computation is infeasible, and the cost of terminating the computation in order to restart it can be prohibitive. Ad hoc methods -- such as inducing a core dump and manually transferring the core map to another program unit -- are not to be trusted. A technique for dynamically reconfigurating software applications is needed, and, as we show, several forms of reconfiguration are desirable for scientific computing. Based upon such motivation, a technology for developing reconfigurable programs is now emerging. The purpose of this paper is to describe these software mechanisms, and then to relate them to potential applications in the scientific computing community.
0 references