README
1Examples
2--------
3Several data type examples and their corresponding parmfiles are available here.
4To test whether migrate works on your machine run simply:
5
6time ../migrate-n parmfile.testml -nomenu
7time ../migrate-n parmfile.testbayes -nomenu
8
9Timing on my Macbook Pro (Intel Core 2 Duo 2.5 GHz with 2 cores, 6 MB L2 Cache,
104 GB memory, 800 MH bus speed) running parmfile.testml and parmfile.testbayes
11using: time ../migrate-n parmfile.x -nomenu, reporting real time
12
13Version 3.2 October 2010
14--------------------------------------------------------------------------------
15compilation using: ML Bayes Notes
16make 1m20.465s 3m30.865s -
17make 1m17.018s 2m53.422s GrandCentral
18make thread 2m56.849s 6m39.735s pthreads
19make mpis (3 nodes) 0m51.867s 2m31.029s -
20make mpis (3 nodes) 1m5.488ss 3m0.103s GrandCentral
21--------------------------------------------------------------------------------
22Timing using the FSU HPC cluster:
23make (1 node) 1m51.763s 4m42.985s
24make mpis (11 nodes) 0m23.215s 0m30.281s
25
26On Mac: compiler gcc 4.2; the threaded version uses 5 threads (suboptimal on a 2-core
27machine, the MPI version uses 3 nodes [1 master + 2 worker]), GrandCentral is the Apple
28architecture for fast threading introduced with MacOS 10.6. the make thread
29seems not to work too well compared to the GrandCentral on a Mac, but may be the only way
30to parallelize the heated chains on LINUX (I have not tried any of this on Windows
31[if you are a specialist on windows parallelizing talk to me please]
32
33Compare your runs with the runs labeled:
34outfile-ml-saved
35outfile-bayes-saved
36
37They should look similar, although there might be differences because
38of optimization (compiler dependent) and hardware differences.
39Differences in these (too) short run are due to different
40optimization on different machines, I doubt that it is
41possible to compare between different computer architectures
42but if the program does finish successfully on the test data
43it is likely that the program will work on your data, too.
44
45
46Peter Beerli
47(beerli@fsu.edu)
48May 2010
49
README-old
1Examples
2--------
3Several data type examples and their corresponding parmfiles are available here.
4To test whether migrate works on your machine run simply:
5
6time ../migrate-n parmfile.testml -nomenu
7time ../migrate-n parmfile.testbayes -nomenu
8
9Timing on my Macbook Pro (Intel Core 2 Duo 2.5 GHz with 2 cores, 6 MB L2 Cache,
104 GB memory, 800 MH bus speed) running parmfile.testml and parmfile.testbayes
11using: time ../migrate-n parmfile.x -nomenu, reporting real time
12
13Version 3.2 October 2010
14--------------------------------------------------------------------------------
15compilation using: ML Bayes Notes
16make 1m21.907s 3m38.899s -
17make 1m9.828s 2m57.881s GrandCentral
18make thread 3m19.316s 6m50.137s pthreads
19make mpis (3 nodes) 0m56.998s 2m13.603s -
20make mpis (3 nodes) 1m5.488ss 3m0.103s GrandCentral
21--------------------------------------------------------------------------------
22Timing using the FSU HPC cluster:
23make (1 node) 1m51.763s 4m42.985s
24make mpis (11 nodes) 0m23.215s 0m30.281s
25
26On Mac: compiler gcc 4.1; the threaded version uses 5 threads (suboptimal on a 2-core
27machine, the MPI version uses 3 nodes [1 master + 2 worker]), GrandCentral is the Apple
28architecture for fast threading introduced with MacOS 10.6. the make thread
29seems not to work too well compared to the GrandCentral on a Mac, but may be the only way
30to parallelize the heated chains on LINUX (I have not tried any of this on Windows
31[if you are a specialist on windows parallelizing talk to me please]
32
33Compare your runs with the runs labeled:
34outfile-ml-saved
35outfile-bayes-saved
36
37They should look similar, although there might be differences because
38of optimization (compiler dependent) and hardware differences.
39Differences in these (too) short run are due to different
40optimization on different machines, I doubt that it is
41possible to compare between different computer architectures
42but if the program does finish successfully on the test data
43it is likely that the program will work on your data, too.
44
45
46Peter Beerli
47(beerli@fsu.edu)
48May 2010
49