I have made a 3-D 360 degree sliding mesh for a rotor-stator device and am not able to read it into Fluent network parallel.

On reading it into Fluent paralell (single machine - dual processor) I get a warning after building the grid that says:

WARNING: Sliding interface nodes are not encapsulated (for stationary grid). Please make modification in serial solver and repartition before continuing with parallel calculation.

WARNING: Sliding interface is not encapsulated (for moving grid). Please make modification in serial solver and repartition before continuing with parallel calculation.

I do not get any warning like this in the serial solver (In the network parallel solver, the grid won't even distribute). Does anyone know what the warnings mean? Thanks.

Did you de-activate "partition across zones" when you partitioned the grid?

The mesh was partitioned across zones. After deactivating 'partition across zones' and creating two partitions, the mesh read into Fluent paralell (non-network version). It seemed to start calculation ok, however I didn't continue to run to even a single inner iteration because the hard disk was working to hard swapping with memory.

However, the mesh is still not reading into Fluent network parallel. Eventually I am going to have to run on an 8 processor network because of the problem size. Do you have any idea what the problem may be. The following is the output of Fluent when reading the mesh in. I am also wondering why after creating the sliding interface, the mesh now has tetrahedral faces (according to Fluent when it reads the mesh). The mesh was originally made with hexahedrals only. Does creation of the interface make tetrahedrals?

*****************

Loading "g:\fluent.inc\fluent5.4\lib\fl111.dmp" Done.

Welcome to Fluent 5.4.8

Copyright 1999 Fluent Inc.

All Rights Reserved

Loading "g:\fluent.inc\fluent5.4\lib\flprim1188.dmp" Done.

Current fluent usage: 1. karlk@bach1 Wed Feb 21 10:53 bach1

License for fluent expires 12-feb-2002.

Reading "g:\hosts.txt"... Done. Host spawning Node 0 on machine "bach1".

------------------------------------------------------------------------------ ID Comm. Hostname O.S. PID Mach ID HW ID Name ------------------------------------------------------------------------------ host* net bach1 nt 76 0 -1 Fluent Host n0 net bach1 nt 196 0 188 Fluent Node

Node 0 spawning Node 1 on machine "jchristian". C:\WINNT\system32

------------------------------------------------------------------------------ ID Comm. Hostname O.S. PID Mach ID HW ID Name ------------------------------------------------------------------------------ n1 net jchristian nt 360 1 116 Fluent Node host net bach1 nt 76 0 80 Fluent Host n0* net bach1 nt 196 0 -1 Fluent Node

> Reading "| gunzip -c G:\Kevala\New_method\Gambit_files\rotor-stator_paralell.cas.gz"... 1096800 hexahedral cells, zone 7, binary. 938304 hexahedral cells, zone 14, binary. 1096800 cell partition ids, zone 7, 2 partitions, binary. 938304 cell partition ids, zone 14, 2 partitions, binary. 283731 triangular interior faces, zone 1, binary.

39424 quadrilateral interface faces, zone 2, binary.

4284 quadrilateral pressure-outlet faces, zone 3, binary.

21240 mixed wall faces, zone 4, binary. 155318 mixed wall faces, zone 5, binary. 3321228 mixed interior faces, zone 6, binary.

44400 quadrilateral interface faces, zone 9, binary.

4992 quadrilateral velocity-inlet faces, zone 10, binary.

48566 mixed wall faces, zone 11, binary. 104680 mixed wall faces, zone 12, binary. 2851764 mixed interior faces, zone 13, binary.

68286 quadrilateral si-parent-face faces, zone 21, binary.

848 quadrilateral si-parent-face faces, zone 20, binary.

751 quadrilateral si-parent-face faces, zone 19, binary.

743 quadrilateral si-parent-face faces, zone 18, binary.

67897 quadrilateral si-parent-face faces, zone 17, binary.

840 quadrilateral si-parent-face faces, zone 16, binary. 283731 interface face parents, binary. 204858 child and parent pointers, parent zone 21, child zone 13, binary.

2544 child and parent pointers, parent zone 20, child zone 12, binary.

2253 child and parent pointers, parent zone 19, child zone 11, binary.

2229 child and parent pointers, parent zone 18, child zone 5, binary. 207275 child and parent pointers, parent zone 17, child zone 6, binary.

2520 child and parent pointers, parent zone 16, child zone 4, binary. 2303796 nodes, binary. 2303796 node flags, binary. Done.

Parallel variables... Building...

grid,

distributing mesh

parts..,

faces. 0 (mpsystem.c@1169): mpt_read: failed: errno = 10054

0: mpt_read: error: read failed: No error

999999 (mpsystem.c@1169): mpt_read: failed: errno = 10054

999999: mpt_read: error: read failed: No error The Parallel FLUENT process could not be started.

Welcome to the buggy world of sliding interfaces in Fluent. Getting things to run with sliding interfaces is very tricky. I suggest that you contact a support engineer who has experience from running this type of simulations. They should see the problems you have.

You have a hex-grid, that is good - I've had a lot less problems getting those to run. You mentioned 8 processors - this is on the limit of what is possible in Fluent with sliding interfaces. After each time-step Fluent has an algorithm to update the sliding interfaces (some kind of search algorithm to find matching cells I think). The time needed for this "update" grows very quickly with the number of CPUs - I've run a 1.3 million cells stator-rotor-stator case recently and for that I couldn't use more than 6 CPUs - if I use more it starts going slower again. Fluent has found a solution to this problem though and version 6 will have a faster algorithm for the interface update - will make it possible to run efficiently on more than 20 CPUs I hope.

Here are a couple of things you can try:

Before and after you define the sliding interfaces do a "Grid/Check". This somehow seems to fix a bug. Another alternative is to first define the interface as translational, delete it, and then re-create it as rotational (intuitive, isn't it ;-)

Experiment with different partitioning schemes - I've had most luck with partitioning based on x-axis.

Switch to Linux - less bugs compared to NT.

Thanks, will try these things.