Adaptive algorithm selection method (AASM) for dynamic software tuning

Author

Suzaki, Kuniyasu ; Kurita, Takio ; Tanuma, Hitoshi ; Hirano, Satoshi

Author_Institution

Electrotechnical Lab., Ibaraki, Japan

fYear

1993

fDate

1-5 Nov 1993

Firstpage

248

Lastpage

256

Abstract

This paper presents a new approach to dynamic software tuning called the adaptive algorithm selection method (AASM). The AASM is built into the calling sequence of a library. When the library is called, the AASM is activated. The AASM selects and executes the optimum algorithm from registered algorithms in a library, based on data and machine type. As a result, the software is automatically tuned and the execution time is shortened. The relation between the data and the best algorithm for a given machine is learned by a neural network from the results of performance tests of the registered algorithms. We experimented on a multi-strings search problem with the AASM on the following machines: the CRAY X-MP/216, FACOM M 1800/30, and SUN Sparc Station 2. From these experiments we demonstrated that the AASM is able to minimize the execution time

Keywords

adaptive systems; dynamic programming; neural nets; software libraries; software performance evaluation; software selection; AASM; CRAY X-MP/216; FACOM M 1800/30; SUN Sparc Station 2; adaptive algorithm selection method; calling sequence; data type; dynamic software tuning; execution time; learning; library; machine type; multi-strings search problem; neural network; optimum algorithm; performance tests; registered algorithms; Adaptive algorithm; Algorithm design and analysis; Artificial intelligence; Laboratories; Neural networks; Operating systems; Search problems; Software libraries; Sun; Testing;

fLanguage

English

Publisher

ieee

Conference_Titel

Computer Software and Applications Conference, 1993. COMPSAC 93. Proceedings., Seventeenth Annual International

Conference_Location

Phoenix, AZ

Print_ISBN

0-8186-4440-0

Type

conf

DOI

10.1109/CMPSAC.1993.404185

Filename

404185