Asynchronous parallel pattern search for nonlinear optimization

Hough, P. D. ; Kolda, T. G. ; Torczon, V. J.

Sandia National Laboratories

关键词: Fault Tolerant Computers;    Fault Tolerance;    Pattern Search;    Computer Architecture;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;   

DOI  :  10.2172/751003
RP-ID  :  SAND2000-8213
RP-ID  :  AC04-94AL85000
RP-ID  :  751003

美国|英语

来源: UNT Digital Library

【 摘 要 】

Parallel pattern search (PPS) can be quite useful for engineering optimization problems characterized by a small number of variables (say 10--50) and by expensive objective function evaluations such as complex simulations that take from minutes to hours to run. However, PPS, which was originally designed for execution on homogeneous and tightly-coupled parallel machine, is not well suited to the more heterogeneous, loosely-coupled, and even fault-prone parallel systems available today. Specifically, PPS is hindered by synchronization penalties and cannot recover in the event of a failure. The authors introduce a new asynchronous and fault tolerant parallel pattern search (AAPS) method and demonstrate its effectiveness on both simple test problems as well as some engineering optimization problems