DRI Terminology (much borrowed from MPI 1.1, propose incorporation into DRI glossary and API spec)

• MPI 1.1 discussion of some of these terms in section 2.3 ( http://www.mpi-forum.org/docs/mpi-11-html/node11.html#Node11 )
• Blocking function
• one that, after it returns, allows the caller to immediately reuse those resources that were provided as parameters
• Non-blocking function
• one that, after it returns, does not allow the caller to immediately reuse those resources that were provided as parameters
• sometimes is accompanied by a corresponding "test" function to test completion of the operation
• NOTE: A non-blocking DRI function does NOT necessarily guarantee an immediate return to the caller. It is possible that a non-blocking function will need to wait for other resources (not user-visible) to become available before proceeding. To provide a more immediate behavior, an implementation may choose to maintain a queue of such operations, and execute them in an independently scheduled execution context (e.g., a separate thread of control). An example of resources not visible to the caller are memory buffers associated with processes on the other "side" of a data reorganization.
• Immediate function:
• one that returns to the caller in timely manner, by not blocking on any user-visible or internal resources before returning
• consequently, this may involve a simple queueing of the request (if the operation cannot be completed without blocking), and orchestrating deferred execution of that request in a separate execution context
  
• may have a corresponding test function so the user can query whether the underlying operation has completed
• Local function
• one that requires no communication with any other DRI process in order to complete
• Non-Local function
• one that may require communication with one or more other DRI processes to complete
• coordinations with other DRI processes may be deemed synchronous or asynchronous
• Synchronous function
• one that cannot complete until another independent event (perhaps completion of a different DRI function call) has been performed (usually by a group of DRI processes)
• In DRI, we need to understand [a]synchronous semantics with respect to the source and destination process groups (there may be 2 groups involved, as in a pipeline design).
• Asynchronous function
• one that involves communication with other DRI processes, but does not require any active coordination in order for the function call to complete.
• Like synchronous functions, need to discriminate between source and destination side process groups in DRI
  
• Example: DRI_Reorg_create might be implemented to publish the details of one "side" of the data reorg being created into a central repository. This could be done (in one sample implementation) by individual processes on one side of the reorg, without any special synchronous coordination.
• Barrier Synchronous function
• one that is scoped to a group of DRI processes, and requires all of those processes to call this SAME function before any of the processes may complete execution of the function