Hi.
I have developed a program that minimizes the free energy in GL models. I am not going to describe my algorithm in any depth, but basically, I store values of certain functions in multi - arrays and then minimize a functional by varying function values in a single point at a time.
Right now, I am doing this on a desktop computer, but I could probably use a cluster for this purpose, and therefore I need do modify my program. I am a physics student and not a programmer and so I don't know a great deal of c++ / multi core programing.
The algorithm I use is certainly threadable. I would do something like varying the value of many points at the same time. If the points are suitably chosen (sufficiently separated), then these processes are completely independent.
Some questions come in to my mind though.
Do I have to tell the compiler in some manner that I want to thread my application?
Since we are talking about a cluster, I assume there is some latency when data travel between machines? Then I assume one has to write some sort of routine that shuffles data between the machines at certain steps during the iteration?
Finally, is this dependent on operative system? I am developing this on a mac, but will probably run it on a linux cluster...
I have developed a program that minimizes the free energy in GL models. I am not going to describe my algorithm in any depth, but basically, I store values of certain functions in multi - arrays and then minimize a functional by varying function values in a single point at a time.
Right now, I am doing this on a desktop computer, but I could probably use a cluster for this purpose, and therefore I need do modify my program. I am a physics student and not a programmer and so I don't know a great deal of c++ / multi core programing.
The algorithm I use is certainly threadable. I would do something like varying the value of many points at the same time. If the points are suitably chosen (sufficiently separated), then these processes are completely independent.
Some questions come in to my mind though.
Do I have to tell the compiler in some manner that I want to thread my application?
Since we are talking about a cluster, I assume there is some latency when data travel between machines? Then I assume one has to write some sort of routine that shuffles data between the machines at certain steps during the iteration?
Finally, is this dependent on operative system? I am developing this on a mac, but will probably run it on a linux cluster...
