Re: [Libmesh-devel] Odd ParametisPartitioner hangups on UB CCR cluster

[Boris B. <bor...@bu...>]

Intel MPI Library Version 2017.0.1

I tried devel mode earlier and it was getting stuck at the same spot inside
the communicate_bins() call; specifically when trying this->comm().get() in
the MPI_Gatherv which was inlined away. I saw no output indicating
something was wrong in terms of asserts, but Ill try dbg and see what else
I can dig up and ill report back.

On Mon, Sep 18, 2017 at 12:54 PM, Roy Stogner <roy...@ic...>
wrote:

>
> On Mon, 18 Sep 2017, Boris Boutkov wrote:
>
>  - I often attach a gdb session to the running program and notice
>> the commonly recurring stack (see below). It seems the issue is
>> always around the HilbertIndices parallel sort communicate_bins()
>> with invalid looking communicator ids in the above PMPI_Allgather
>> calls.
>>
>
> If it's replicable for particular mesh sizes, could you try running
> in devel (or better, dbg) modes and see if you get any more
> informative output?
>
> What MPI implementation+version are you using?
> ---
> Roy