It's just that providing a "more meaningful" key can help when debugging or managing through the OS. You still have to generate some sort of "sequence" number per-communicator, as successive allocations on the same communicator must return different memory. So, communicator alone is not enough to uniquely identify/select an allocation. Implementations will have to generate a "key" to pass to the OS, and so all ranks must be able to generate identical keys for corresponding calls.

_______________________________________________
Douglas Miller BlueGene Messaging Development
IBM Corp., Rochester, MN USA Bldg 030-2 A410
dougmill@us.ibm.com Douglas Miller/Rochester/IBM

Pavan Balaji ---12/07/2010 08:08:55 AM---On 12/07/2010 07:58 AM, Douglas Miller wrote:

Pavan Balaji <balaji@mcs.anl.gov>
12/07/2010 08:08 AM

To

mpi3-hybridpm@lists.mpi-forum.org

cc

Douglas Miller/Rochester/IBM@IBMUS

Subject

Re: [Mpi3-hybridpm] Hybrid WG meeting at the Forum

On 12/07/2010 07:58 AM, Douglas Miller wrote: > In order to ensure all nodes get the same memory segment, they all have > to use the same "key". At least if this is implemented on Unix SysV or > POSIX shmem. Then there's the possibility of some nodes allocing on a > sub-comm first, etc. I think its just avoiding a lot of potential > complications and don't see what the downside is. In that sense, isn't the communicator a "key"? -- Pavan -- Pavan Balajihttp://www.mcs.anl.gov/~balaji