Master/Slave

[miguelarriaga]

Hi,

I'm trying to set some Master/Slave conditions and that information is not being passed to the Parallel input. I'm using a 4-node example with 2 displcement dof and one temperature. I want to make the X displacement of node 2 a slave dof of the X displacement of node 3.

4 -------- 3
|          |
di         |
|          |
1 -- di -- 2

MASTer  !Mx,My,Sx,Sy,rl(x,y,t)
  Node   di,di,di, 0,   0,1,1


In the initial serial set-up run, if I plot the id31 array, before the uoutdom subroutine in pmacr7 I have the following:

Node	DispX	DispY	Temp
1	-1	1	2
2	5	3	4
3	5	6	7
4	-17	8	9

As we can see, the displacement on the X direction has the same numbering for nodes 2 and 3, which means the Master/Slave routine was successfull.

However, after uoutdom the array id31 becomes:

Node	Dx	Dy	T
1	0	1	2
2	3	4	5
3	6	7	8
4	0	9	10

which is the same matrix that is exported to the "EQUAtion numbers" list in the parallel input if OUTD,UNBL is used (only using one processor). Here the array no longer shows the same number in DX2 and DX3. If I print out this array right before the set-up run finishes I get the first array again, which means that id31 was fixed after the uoutdom changed it.

Finally, if I print that same array on the parallel run right before usolve I see that the array is now:

Node	Dx	Dy	T
1	-1	1	2
2	3	4	5
3	6	7	8
4	-17	9	10

Is this behaviour normal? When does the parallel run generate id31 and how can I fix it to include the desired Master/Slave behaviour?

Thank you,
Miguel

[miguelarriaga]

PS1: I'm using version 8.3 (Switching to 8.4 would be a last resort solution because the code has been heavily patched)
PS2: The number -17 is because I'm actually using a more complicated element with 5 masked dof.

[FEAP_Admin]

8.3 does not have parallel contact.  you will need 8.4.

[miguelarriaga]

I was under the impression that the function MASTer was under the scope of Rigid Body Analysis for Small Displacements

[FEAP_Admin]

I misunderstood.  I thought you were trying to do contact.

Notwithstanding, what you are trying to do in not part of parFEAP (not even version 8.4).
If you want to do this in parallel then you would have to adjust the graph partitioning to
properly handle it.  We do have this on our to do list but have not done as of yet.

[miguelarriaga]

Thank you for your answer.
Could you explain to me what is the problem with implementing this? I was thinking that something like this could work:

At the Set-up run, for each partition, loop over the nodes of that partition and get the master nodes with mr(np(100))
If the master nodes are not on that partition, then add them as Ghost nodes and create a local array equivalent to mr(np(100))
Finally make uoutdom export a LINK command for all local master/slave pairs.

Would this work?