[ HPF Home | Versions | Compilers | Projects | Publications | Applications | Benchmarks | Events | Contact ]

---

Markov Chain Monte Carlo (MCMC) for carcinogenesis models

Back to Applications

g@trex.fhcrc.org

Maximum likelihood calculations, Markov Chain Monte Carlo (MCMC)

Number of Lines of Code: ~500 (program with main loop is linked to a packag

Target Platforms and HPF Compilers Used:

LINUX for Intel

Coding Styles (data decompositions, computational methods):

F90, trivial parallelization via distribution of data across procs. Indep. numerical evaluation of individual records. Use of non-pure subroutines and functions within main loop can cause erroneous results.

Extrinsic Interfaces Used (and reasons):

for data read HPF_SERIAL, otherwise program freezes when 2 or more procs are used. Same for data write, and for operations that use system() calls.

Performance Information, if Available (including any possible comparisons to MPI and/or OpenMP):

Our MCMC calculations have x6.5 speedup on 8 processors (4 dual PENTIUMII machines)
over single procs. No comparisons with MPI yet.

Please comment on any aspects of the application that might be interesting, including any problems using HPF effectively:

It would help if some compiler warnings concerning potential problems of parallization were more specific. A publication of potential problems, and case studies of generic numerical problems solved via HPF may be very useful.

URL

http://biostat1.fhcrc.org