Distributed optimization of linear programs using Dantzig-Wolfe decomposition
Optimization of black-angular linear programs using distributed computers is studied. The approach is based on the Dantzig-Wolfe decomposition algorithm. In a Dantzig-Wolfe decomposition, a block-angular linear program with R blocks is decomposed into R + 1 semi-autonomous processes. One of them is the coordinating process which generates prices, and the remaining R processes will respond to the prices through the submission of proposals. Since these processes can be executed concurrently and asynchronously, the control of the flow of prices and proposals information among the processes is critical to the rate the entire system achieves optimality. How does the control of information flow affect the dynamics of the concurrent process An information scheme is a distributed Dantzig-Wolfe decomposition controls the timing of the availability and utilization of the prices and proposals information. Focus is on four information control schemes: the Basic Information Scheme; the Early Termination Information Scheme; the Early Stat Information Scheme; and the Intermediate Prices Information Scheme. The efficiency, load-balance, and communication requirements of using distributed Dantzig-Wolfe decomposition to solve block-angular LP's are studied.
- Research Organization:
- Tennessee Univ., Knoxville, TN (USA)
- OSTI ID:
- 6988042
- Country of Publication:
- United States
- Language:
- English
Similar Records
Decomposition of linear programs using parallel computation. Technical report
Decomposition of linear programs using parallel computation. Technical report