Hi Xin, > Memory pressure is the reason I turn to step-40. What I'm trying to > figure out is that: instead of one layer of ghost cells, is it possible > to have multiple layer ghost cells within the framework of step-40? > Besides the information (solution) on the adjacent cells on the neighbor > processor, my application probably asks for information on other cells > owned by the neighbor processor, but not immediately adjacent to any > cell owned by the current processor. The other way to see this is that > my application may require information on all active cells within a > coarser level cell that is not owned by the current processor, but > adjacent to cells owned by the current processor.
Suppose we have a situation like this where cell 0 is on the current processor: _______________ |1|_| | | |2|_|___| | | | | 0 | |___|___|_______| Are you implying that you'd also need cells 1 and 2 (or their children if they are refined in turn)? Due to the recursivity the communication volume may become arbitrarily large. The other, better behaved option would be to provide two layers of ghost cells, that is all cells except 1 and 2. Would this be sufficient? Best regards, Carsten > Thank you. > Best regards, > Xin > > > > > On Tue, Nov 29, 2011 at 8:26 PM, Wolfgang Bangerth > <[email protected] <mailto:[email protected]>> wrote: > > > > I'd like to use the nice dealii framework of parallel computing with > > multiple processors using distributed memory, which is illustrated in > > step-40. It is clear that we can recover the solution on the processor > > owned cells as well as on a layer of ghost cells. My question is > whether > > dealii has some functionalities to reconstruct the solution on the > entire > > distributed triangulation, including artificial cells, > > The information is of course there, distributed across all > processors. But > it's not on every single processor, for a simple reason: the > framework is > designed for the case where no single processor has enough memory to > store the > entire solution. The existence of artificial cells is one case in point: > artificial cells on one processor are not necessarily locally-owned > cells on > another processor; in general, the global mesh will consist of the > children/grand children/... of artificial cells on one processor. > > If you want to work in a framework where every processor has access > to all the > information, take a look at steps 17 and 18. > > All of this aside: If you tell us what you want to do, maybe we can > come up > with a better solution! > > Best > W. > > ------------------------------------------------------------------------- > Wolfgang Bangerth email: > [email protected] <mailto:[email protected]> > www: http://www.math.tamu.edu/~bangerth/ > > > > > > _______________________________________________ > dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
