Performance feature identification by comparative trace analysis
This paper is concerned with optimizing ®nancial programs through the use of performance models to promote the best use of available computational resources. The work necessarily focuses on run-time optimization, where a performance model of the program is combined with knowledge of run-time conditions (such as input data and system state) to produce information useful in the optimization process. Two speci®c techniques are presented: a method for dynamically selecting the best algorithm for pricing a particular ®nancial product based on run-time performance data (optimization at the single calculation level), and a method of using performance data with heuristic techniques for run-time scheduling of a large number of option-pricing calculations (both sequential and parallel) over the entire computer system. Demonstrations of these techniques for example calculations and resources typically used in ®nancial institutions are also described. Ó 2000 Elsevier Science B.V. All rights reserved.