Hello,

While using GARD on our cluster, for some genes with many taxa, the computation takes too much time, and the job is finally killed by the grid engine queuing system. At that point I have the choice between 1) running it with more processors, 2) running it on a longer queue, or 3) modify the dataset. Regarding solution 1 and 2, the administrator of our cluster asked me the following questions that I was not able to answer...:

1. Is there any way to restart a calculation from where it ended? Does
the program produce any kind of checkpoint information that could be
used to restart it? If it did, any queue time limitations would not
matter, because you could always continue the run as a separate job.

2. What is the parallel efficiency of HYPHY, i.e., if you run it (with
the same input) on N processors, and then on 2N processors, is the
second run going to be twice shorter than the first one, or only a
little shorter?

Thanks!

-Tristan

Dear Tristan,

To answer your questions:

1). The current implementation does not generate checkpoints; but it is not difficult to modify to do so. Are you running into problems with unfinished runs?

2). A GA does not run faster on more processors; but you can increase population size (e.g. one individual per processor) and speed up convergence that way, although the exact impact is difficult to assess.

Cheers,
Sergei

Quote:

1). The current implementation does not generate checkpoints; but it is not difficult to modify to do so. Are you running into problems with unfinished runs?

Yes, and we were wondering if we can restart the GA from where it was stopped.

-Tristan