I have a model with about 1 million DOF and I solve for about 1000 modes using QRdamp. I found that the analysis was too slow and took too much scratch space. Is this a system issue or a solver issue?
The QRdamp solver was designed to solve for a small number of modes (e.g., about 50), and is not optimized for speed when a large number of modes are requested. For this problem, the Block Lanczos portion of the solve used only 1252 MB of memory at any time; for the remaining portion of the solve (QRdamp code), it seems as though about 40 GB were required. If your system does not have >40 GB of memory, then a certain amount of the solution will have to be done out-of-core. If the memory in the system is <<40 GB, then a significant amount of virtual memory may be swapped in-and-out from the disk, making the process very slow. Significant improvements to this portion of the code have been made for ANSYS v12. |
||
|