On Thu, 19 Nov 2009, Vijay S. Mahadevan wrote:

> I think Hilbert library should be enabled by default if someone
> wants to run in parallel.

You saw the log message about disabling libHilbert, but I guess you
missed the discussion about why?

It was giving us collisions.  Two separate users managed to find cases
where ~10 levels of refinement would cause libHilbert to be unable to
distinguish between two nodes, leading to an I/O assertion failure in
debug modes or a corrupt output file in optimized modes.  We're not
turning it back on until that's fixed.

But of course we don't want to disable MPI for everyone either.  I've
written a tolerable default global_indices(), and I'm testing it now.

The testing will be unfortunately less than such a low lying change
deserves, but better to have code that might not work than code that's
definitely been found to have failure cases...
---
Roy

> Or we should disable MPI in configure if
> Hilbert is
> disabled. At least a error message there will avoid further problems
> for now until we have a default implementation for global_indices().
>
> On Thu, Nov 19, 2009 at 4:57 PM, Vijay S. Mahadevan <[email protected]> wrote:
>> I think the problem is because of disabling Hilbert library. In fact
>> nothing in mesh_communication_global_indices.C is implemented at all
>> if this library is disabled. But I am curious on how the global
>> indices got done if only one processor is used !
>>
>> On Thu, Nov 19, 2009 at 4:49 PM, Kirk, Benjamin (JSC-EG311)
>> <[email protected]> wrote:
>>>> I saw in the log file that recently Roy disabled libHilbert...
>>>
>>> Ah, that's what I get for falling behind on my reading.
>>>
>>>
>>
>
> ------------------------------------------------------------------------------
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
> trial. Simplify your report design, integration and deployment - and focus on
> what you do best, core application coding. Discover what's new with
> Crystal Reports now.  http://p.sf.net/sfu/bobj-july
> _______________________________________________
> Libmesh-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/libmesh-devel
>

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Libmesh-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-devel

Reply via email to