Dear Martin,
This is more reasonable. The keypoint here is
DDRMM parameter (Dynamic Data Recovery Matrix Method). In the case of dynamic analysis with the modal formulation, two methods of data recovery are available. The default procedure is usually much more efficient for data recovery operations when the modal formulation is used for dynamic analysis. In this procedure, the dependent components of the eigenvectors used in the modal formulation are first recovered by SDR1 using the same sequence of operations described for problems in real eigenvalue analysis (after eliminating unwanted modes). The complete eigenvectors are used as the input to SDR2 in order to determine the requested forces and stresses in terms of modal coordinates.
In dynamic analysis, you may restrict the output request to the response of selected points in the solution set. In this case, the time-consuming operations in the recovery of the dependent degrees-of-freedom can be avoided. This solution set processing is performed by VDR (Vector Data Recovery).
The modal data generated by SDR1 and SDR2 are formed into a data matrix by
DDRMM. The output quantities are determined by multiplying the data matrix by the modal solution vectors. The computer time for this matrix multiplication is proportional to the number of output times, the number of modes used in the modal formulation, and the number of output frequencies or output time steps. If both the number of output items and the number of modes are small, the computer time required will not be significant when compared to the total problem solution time. Although the DDRMM method of data recovery is more efficient, a complete {ug} is not formed. For this reason the method does not permit the creation of deformed structure plots, grid point force balance or the use of the mode acceleration technique.
The alternate procedure may be invoked using PARAM,DDRMM,-1. With this method the solution vectors are first transformed from modal to physical coordinates. PARAM,DDRMM,-1 and PARAM,CURVPLOT,1 are required in the Bulk Data to create structure plots at specified frequencies. (Note that PARAM,DDRMM,-1 generally increases the amount of computer time and is not recommended unless otherwise required.)
Regarding RAM performance, I have learned from NX NASTRAN developers that Nastran is very much file based and IO performance is almost always the limiting performance factor. Hardware money is much better spent on disk striping (i.e. RAID 0, modern fast SSD) rather than multiple processors. Additional memory helps to some extent as the modern operating systems (Windows and Linux) use unallocated RAM as IO cache.
This point is critical:
all of the unallocated RAM is used by the OS as cache to help IO. If you take a look to the TASK MANAGER you will see that NX NASTRAN use the RAM memory you assigned by MEM command, ie, if you set MEM=5GB then NX NASTRAN will use 5 GB RAM as maximum, the rest of RAM will be used by the OS as IO cache, OK?.
Then, if in a dynamic response analysis you get any message in the output file *.F06 like:
Code:
AVAILABLE MEMORY IS LESS THAN THE MEMORY NEEDED TO HOLD
LOADS AT DIFFERENT FRQUENCIES/TIMES INTERVALS. THIS WILL
IMPACT PEFORMANCE AS SPILL LOGIC WILL BE USED. PLEASE
INCREASE MEMORY BY xxxxxxx WORDS TO HOLD THE LOADS INCORE
then simply increase the RAM memory devoted to the solver. If the NX NASTRAN solver ask more than 8 GB RAM, then you will need to use the ILP-x64 NX NASTRAN solver (with 8 bytes per word) instead the regular LP-64 solver (by default).
Best regards,
Blas.
~~~~~~~~~~~~~~~~~~~~~~
Blas Molero Hidalgo
Ingeniero Industrial
Director
IBERISA
48011 BILBAO (SPAIN)
WEB:
Blog de FEMAP & NX Nastran:
