Branch-and-Bound based Algorithms on Parallel Computing
Environments
-
General Backtracking methods
- Combinatorial problem(s) involved in the implementation:
0/1-knapsack problem, 15-puzzle, search for Golomb rulers,
search for transition tables of cellular automata, and
game tree search.
- The motivation for parallelism:
1. Backtracking is time consuming. 2. In principle easy to parallelize.
3. Load balancing is an interesting challenge.
- The type of hardware and number of processors used:
MIMD computers from workstation networks through
machines with 1024 processors: transputers, IBM SP.
SIMD computer MasPar MP-1 with 16384 processors.
- Often almost linear speedup over the non-parallel
version.
- Availability of the software:
The reusable library PIGSeL is available upon request.
- More information is available at
this page
or contact Peter Sanders .
- Address: University of Karlsruhe, Dept. of Computer Science,
Chair for Informatics in Science and Engineering, Germany.
- General (Mixed) Integer Programming:
- Combinatorial problems(s) involved in the implementation:
unknown.
- The motivation fro parallelism:
Speed-up in commercial problems, i.e. reducing times to solve some
problems from perhaps overnight to being able to do a couple of
runs per day. Important in scenario analysis in many industries.
- Type of hardware used:
UNIX workstations + Windows 3.1/Win95/WinNT PCs.
- How results compare with the non-parallel version:
Machines 4 x RS6K/390s
- Model Serial Time Computers Parallel Time Speed Up
- SPLY 4440 4 x 390 540 8.22
- BPOR 3545 4 x 390 461 7.69
- NPSOX 660 4 x 390 245 2.69
- Slaves Elapsed Time (secs) Speedup
- 0 (serial) 515 1.00
- 1 552 0.93
- 2 319 1.61
- 3 232 2.22
- 4 181 2.85
- 5 135 3.81
- 6 124 4.15
- The software is commercial.
- To get more infor about it check this
page , or contact: Bob Daniel, Dash Associates, Blisworth House,
Blisworth, Northants NN7 3BX, UK.
- A parallel synchronized Branch and Bound.