253. Memory Model

Up: Contents Next: Synchronization Calls Previous: Request-based RMA Communication Operations

The memory semantics of RMA are best understood by using the concept of public and private window copies. We assume that systems have a public memory region that is addressable by all processes (e.g., the shared memory in shared memory machines or the exposed main memory in distributed memory machines). In addition, most machines have fast private buffers (e.g., transparent caches or explicit communication buffers) local to each process where copies of data elements from the main memory can be stored for faster access. Such buffers are either coherent, i.e., all updates to main memory are reflected in all private copies consistently, or non-coherent, i.e., conflicting accesses to main memory need to be synchronized and updated in all private copies explicitly. Coherent systems allow direct updates to remote memory without any participation of the remote side. Non-coherent systems, however, need to call RMA functions in order to reflect updates to the public window in their private memory. Thus, in coherent memory, the public and the private window are identical while they remain logically separate in the non-coherent case. MPI thus differentiates between two memory models called RMA unified, if public and private window are logically identical, and RMA separate, otherwise.

In the RMA separate model, there is only one instance of each variable in process memory, but a distinct public copy of the variable for each window that contains it. A load accesses the instance in process memory (this includes MPI sends). A local store accesses and updates the instance in process memory (this includes MPI receives), but the update may affect other public copies of the same locations. A get on a window accesses the public copy of that window. A put or accumulate on a window accesses and updates the public copy of that window, but the update may affect the private copy of the same locations in process memory, and public copies of other overlapping windows. This is illustrated in Figure 20 .

Image file

Figure 20: Schematic description of the public/private window operations in the  MPI_WIN_SEPARATE memory model for two overlapping windows.

In the RMA unified model, public and private copies are identical and updates via put or accumulate calls are eventually observed by load operations without additional RMA calls. A store access to a window is eventually visible to remote get or accumulate calls without additional RMA calls. These stronger semantics of the RMA unified model allow the user to omit some synchronization calls and potentially improve performance.

Advice to users.

If accesses in the RMA unified model are not synchronized (with locks or flushes, see Section Lock ), load and store operations might observe changes to the memory while they are in progress. The order in which data is written is not specified unless further synchronization is used. This might lead to inconsistent views on memory and programs that assume that a transfer is complete by only checking parts of the message are erroneous. ( End of advice to users.)
The memory model for a particular RMA window can be determined by accessing the attribute MPI_WIN_MODEL. If the memory model is the unified model, the value of this attribute is MPI_WIN_UNIFIED; otherwise, the value is MPI_WIN_SEPARATE.

Up: Contents Next: Synchronization Calls Previous: Request-based RMA Communication Operations

Return to MPI-3.1 Standard Index
Return to MPI Forum Home Page

(Unofficial) MPI-3.1 of June 4, 2015
HTML Generated on June 4, 2015