Did you CC the boost list? You might want to ask them this question. Plus, we can't answer any questions about the Microsoft HPC/MPI implementation -- this list is for support of the Open MPI software package, not Microsoft's. Sorry.
I suspect that it probably has something to do with how much work you give to boost to do. Parallelization generally does best when you throw enough work at it that overcomes the overhead of parallelization (e.g., network latency and bandwidth, etc.).
On Aug 16, 2011, at 11:59 PM, Chen Kai wrote:
> Environment: Boost1.47 + Microsoft HPC 2008 SDK
> I find Delta Stepping Dijkstra and Crauser Dijkstra in boost, and both of them are slower than sequential dijkstra in boost. (About 10 times slower. I used 10 graphs, the biggest graph contains 10 million nodes and 10 million edges)
> Q1: I wonder why the parallel dijkstra is slower than sequential version?
> Q2: Can I adjust the number of threads that Delta Stepping used? (For exemple, I want it use 1, 2, 3, 4 threads respectively. And Can I make it sequential version?)
> Chen Kai
> School of Computer Science, Fudan University
> blog: http://www.cnblogs.com/remlostime/
> users mailing list
For corporate legal information go to: