Hi to all,
I am planning on running some large 2D Bosons simulations with ALPS in a cluster. Specifically I will we using the worm algorithm(and dir_loop for some checks). The cluster I am running my simulations in uses a PBS queuing system. All the tutorials and examples I have seen online are for single machine or assuming one doesn't have to use a queuing system. Is there any PBS functionality implemented in pyalps?
Also, and this is a practical question, for running large simulations with ALPS, I understand from the examples and documentation that the best way to do it is to set a very large number of sweep steps and then run it for a fixed time. After that time check if the measurements I am interested in have converged and if not, run it for a bit longer. Is that the best option or is there a better way of doing it?
Thanks in advance,
Francisco Cordobes
Hi Francisco,
We have tutorials that show the use of the MPI version of the ALPS codes. Just add the --mpi option to the command line and the Monte Carlo codes run nicely in a parallel way. You don't need anything special for integration with PBS, except that you should remember to set a time limit that is shorter than the time limit of the queue so that the code can still write a checkpoint.
Tama Ma pingnang@phys.ethz.ch is finishing a new and much faster worm code and you might want to get in touch with him regarding that code.
Matthias
On Mar 19, 2013, at 2:40 AM, Francisco Cordobés ghiret@gmail.com wrote:
Hi to all,
I am planning on running some large 2D Bosons simulations with ALPS in a cluster. Specifically I will we using the worm algorithm(and dir_loop for some checks). The cluster I am running my simulations in uses a PBS queuing system. All the tutorials and examples I have seen online are for single machine or assuming one doesn't have to use a queuing system. Is there any PBS functionality implemented in pyalps?
Also, and this is a practical question, for running large simulations with ALPS, I understand from the examples and documentation that the best way to do it is to set a very large number of sweep steps and then run it for a fixed time. After that time check if the measurements I am interested in have converged and if not, run it for a bit longer. Is that the best option or is there a better way of doing it?
Thanks in advance,
Francisco Cordobes
Hi,
Thanks.
I have managed to get the bits you mention working. I was just wondering if pyalps can also generate automatically PBS queuing files too as so far I have been doing it by hand. I can always adapt my own python code for that, but if it's already in pyalps, it would be quicker.
Thanks. Francisco Cordobes
On 19 March 2013 16:49, Matthias Troyer troyer@phys.ethz.ch wrote:
Hi Francisco,
We have tutorials that show the use of the MPI version of the ALPS codes. Just add the --mpi option to the command line and the Monte Carlo codes run nicely in a parallel way. You don't need anything special for integration with PBS, except that you should remember to set a time limit that is shorter than the time limit of the queue so that the code can still write a checkpoint.
Tama Ma pingnang@phys.ethz.ch is finishing a new and much faster worm code and you might want to get in touch with him regarding that code.
Matthias
On Mar 19, 2013, at 2:40 AM, Francisco Cordobés ghiret@gmail.com wrote:
Hi to all,
I am planning on running some large 2D Bosons simulations with ALPS in a cluster. Specifically I will we using the worm algorithm(and dir_loop for some checks). The cluster I am running my simulations in uses a PBS queuing system. All the tutorials and examples I have seen online are for single machine or assuming one doesn't have to use a queuing system. Is there any PBS functionality implemented in pyalps?
Also, and this is a practical question, for running large simulations with ALPS, I understand from the examples and documentation that the best way to do it is to set a very large number of sweep steps and then run it for a fixed time. After that time check if the measurements I am interested in have converged and if not, run it for a bit longer. Is that the best option or is there a better way of doing it?
Thanks in advance,
Francisco Cordobes
No, but it should be easy to write a Python function to do that. Feel free to post such functions on the mailing list if you decide to write them
Matthias
On Mar 19, 2013, at 7:15 PM, Francisco Cordobés ghiret@gmail.com wrote:
Hi,
Thanks.
I have managed to get the bits you mention working. I was just wondering if pyalps can also generate automatically PBS queuing files too as so far I have been doing it by hand. I can always adapt my own python code for that, but if it's already in pyalps, it would be quicker.
Thanks. Francisco Cordobes
On 19 March 2013 16:49, Matthias Troyer troyer@phys.ethz.ch wrote:
Hi Francisco,
We have tutorials that show the use of the MPI version of the ALPS codes. Just add the --mpi option to the command line and the Monte Carlo codes run nicely in a parallel way. You don't need anything special for integration with PBS, except that you should remember to set a time limit that is shorter than the time limit of the queue so that the code can still write a checkpoint.
Tama Ma pingnang@phys.ethz.ch is finishing a new and much faster worm code and you might want to get in touch with him regarding that code.
Matthias
On Mar 19, 2013, at 2:40 AM, Francisco Cordobés ghiret@gmail.com wrote:
Hi to all,
I am planning on running some large 2D Bosons simulations with ALPS in a cluster. Specifically I will we using the worm algorithm(and dir_loop for some checks). The cluster I am running my simulations in uses a PBS queuing system. All the tutorials and examples I have seen online are for single machine or assuming one doesn't have to use a queuing system. Is there any PBS functionality implemented in pyalps?
Also, and this is a practical question, for running large simulations with ALPS, I understand from the examples and documentation that the best way to do it is to set a very large number of sweep steps and then run it for a fixed time. After that time check if the measurements I am interested in have converged and if not, run it for a bit longer. Is that the best option or is there a better way of doing it?
Thanks in advance,
Francisco Cordobes
comp-phys-alps-users@lists.phys.ethz.ch