Computation cost, space and Solver error definitions.

apadzak

Hi Victor,

I have failed few times one simulation few times due to lack of resources. Then I redirected Matrix to a file, but "Assembling matrix" process which is 2 steps before it happens, still uses system directory space to dump some temporary data there. Is there any option that Mecway will not use system space or redirecting all standard output information to the simulation directory?

I also got an error message from solver and I am just wondering if there is any list of Codes that could be available to users so we can understand the mistake and correct it if possible, before raising issue to you?

Thanks

apadzak

I've just spoted that Solving matrix also uses system space.

Victor

Changing the temporary folder

There's an undocumented setting to change the temporary folder which is used by the solver and mesher. This can be useful if the default drive is full. Edit the configuration file which is located in

%localappdata%\Mecway_Limited\ ... \user.config

Insert a new setting next to the other ones like

            <setting name="TemporaryFolder" serializeAs="String">

                <value>C:\users\YOURUSERNAME\desktop\mw</value>

            </setting>

but replace the path with one you prefer.

Then start Mecway.

Victor

Error codes

1-4 digit error codes without a P in front are from ARPACK when solving large eigenvalue problems. From the ARPACK manual:

          =  1: Maximum number of iterations taken.

                All possible eigenvalues of OP has been found. IPARAM(5)  

                returns the number of wanted converged Ritz values.

          =  2: No longer an informational error. Deprecated starting

                with release 2 of ARPACK.

          =  3: No shifts could be applied during a cycle of the 

                Implicitly restarted Arnoldi iteration. One possibility 

                is to increase the size of NCV relative to NEV. 

                See remark 4 below.

          = -1: N must be positive.

          = -2: NEV must be positive.

          = -3: NCV must be greater than NEV and less than or equal to N.

          = -4: The maximum number of Arnoldi update iterations allowed

                must be greater than zero.

          = -5: WHICH must be one of 'LM', 'SM', 'LA', 'SA' or 'BE'.

          = -6: BMAT must be one of 'I' or 'G'.

          = -7: Length of private work array WORKL is not sufficient.

          = -8: Error return from trid. eigenvalue calculation;

                Informational error from LAPACK routine dsteqr.

          = -9: Starting vector is zero.

          = -10: IPARAM(7) must be 1,2,3,4,5.

          = -11: IPARAM(7) = 1 and BMAT = 'G' are incompatable.

          = -12: IPARAM(1) must be equal to 0 or 1.

          = -13: NEV and WHICH = 'BE' are incompatable.

          = -9999: Could not build an Arnoldi factorization.

                   IPARAM(5) returns the size of the current Arnoldi

                   factorization. The user is advised to check that

                   enough workspace and array storage has been allocated.

Error codes beginning with "P" are from the Pardiso linear equation solver:

P-1 Input inconsistent

P-2 Not enough memory

P-3 Reordering problem

P-4 Zero pivot, numerical factorization or iterative refinement problem

P-5 Unclassified (internal) error

P-6 Reordering failed (matrix types 11 and 13 only)

P-7 Diagonal matrix is singular

P-8 32-bit integer overflow problem

P-9 Not enough memory for OOC

P-10 Problems with opening OOC temporary files

P-11 Read/write problems with the OOC data file

6-digit error codes beginning with "E" are bugs and there's no list of the meanings of the codes. Please report any of these you find.

apadzak

Thank you Victor,

Mine error was P-3. Sounds like matrix issue, sufficient space on system drive got rid of the problem.

I will try to use above description for size problems.
 
Thanks.

[Deleted User]

Very useful information Victor. This should probably be in the user manual.

Sergio

Nice!

One thing that I notice while running insane big models in CCX is that Mecway itself took a lot of memory also while is being solved. One workaround is to export the model as .inp and then close Mecway and lunch CCX from command line, then we can open the .frd from Mecway also... but problem arise if the model have more than one component, the posibility of posptrocess each one individualy (hide/show) is lost. Is there a way to get back this feature? At least for Mecway's created models, some kind of importing results option.

On the other side, is there another undocumetated feature in the configuration file to hide the origin and show the max resolution in contour plots?

Regards!

Victor

Yea re-importing orphaned frd files is something somebody else wanted too to work around a different issue so I might to to add that.

No options for show origin and resolution but the next version (v6, beta coming within a couple of weeks hopefully) will store the show origin setting.

Regarding including the error codes and/or temporary folder options in the manual, I think they're too hacky for that. It's going to lead to new sets of problems from people struggling with the fix instead of the original problem. A few of the more meaningful P- error codes already display their text descriptions when they appear though.

UGMENTALCASE

Sorry to jump on this thread but I keep getting a p-2 memory error which trying to mesh. I have 32GB of RAM in my laptop and a monitor on it and it's not even passing 20% usage when I get the P-2 error, is there something I can change to help stop this. I happy that this sort of thing runs high on memory, but I'm not seeing that affect.

Thanks in advance

Sergio

Have you tryied with with a bigger size of elements? First order in place of second? Local refining? For big insane meshes I have noted that Salome is better (even if he is using Netgen also). But again, is necesary such a quantity of elements?

UGMENTALCASE

Tried all sorts but now getting the ngmesh.exe failure. Just comes up as stopped working. I'll have to try a number of other things, not sure what though! :-)

Sergio

For some parts Mecway mesher (Netgen) crashes, but if you use Netgen alone works, othewhise try Step/IGS Doctor on Netgen. A third solution is to use Gmsh... and last is use Salome.

Good luck!

Anyway if you have the geometry you could try to simplify on the CAD program.

Victor

For 64 bit Windows with a big enough hard drive and page file, there's no such thing as not enough memory because it can always use more disk space. So it could be that disk space or page file maximum size is too small, or it could be a bug. Make sure it's 64-bit Mecway with Help->About and look at the first line.

How many nodes and elements does the mesh have? It sounds like it might be so high that solving time could become a problem (hours or overnight) even if it does work. Often you can dramatically simplify a model as Sergio suggested - also remove small details. This especially saves time and allows you to do more solves and some mesh refinement.

UGMENTALCASE

Hi Victor,

I'm trying to change the directory with the instructions above, but I get an error each time I open Mecway, it says it needs to reset to default. Is there anywhere specific it needs to go? I've tried an attached USB drive path and a local path and I get the error on both

Cheers

Victor

Please attach the user.config. Maybe it's not correct.

Victor

On second thought, please don't publish a user.config from Mecway on the internet because if you're a paid customer, it contains your product key which is private. Perhaps send this through email instead.

UGMENTALCASE

Good point!

Didn't think of that.

Cheers