Well it is an important clue...
However, as for the startup development, we plan to think it in a sequential calculation way, because the lambda developper will first want to develop and test it in single-processor simulations.
But, of course it will be a compulsory point on which we will need also to work on in the near future.
So, if from now there are some people aware of parallel computing on clusters (with openMP for example) then please join the project !
An OpenFlower Admin.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I can help here as well, being familiar with distributed memory algorithms and coding. One suggestion though and that is that it is fine to prototype a sequential algorithm to see how it fits together, but in my experience the serial algorithms are not easily transformed into a cluster aware sw architecture. It is too easy to make assumptions in the single processor sw structure that will destroy any chances of a successful port. Simple examples are the cost of data scatter and gathers, trivial in the single processor case, the holy grail in cluster computing.
Unfortunately, I have no idea what you guys have in mind what you want to build. So before diving to deep into this point, why don't you describe the basic requirements and product architecture you had in mind.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Since CFD is so computationally demanding, did you include a requirement that the baseline is cluster aware?
Well it is an important clue...
However, as for the startup development, we plan to think it in a sequential calculation way, because the lambda developper will first want to develop and test it in single-processor simulations.
But, of course it will be a compulsory point on which we will need also to work on in the near future.
So, if from now there are some people aware of parallel computing on clusters (with openMP for example) then please join the project !
An OpenFlower Admin.
I can help here as well, being familiar with distributed memory algorithms and coding. One suggestion though and that is that it is fine to prototype a sequential algorithm to see how it fits together, but in my experience the serial algorithms are not easily transformed into a cluster aware sw architecture. It is too easy to make assumptions in the single processor sw structure that will destroy any chances of a successful port. Simple examples are the cost of data scatter and gathers, trivial in the single processor case, the holy grail in cluster computing.
Unfortunately, I have no idea what you guys have in mind what you want to build. So before diving to deep into this point, why don't you describe the basic requirements and product architecture you had in mind.