MPI offers a number of variants of send including the MPI_Rsend operation, which assumes the receiver is ready to receive at the time of sending. What optimizations in implementation are possible if this assumption is correct and what are the repercussions of this assumption being false?
What will be an ideal response?
The key optimization is that the sender does not have to check that the receiver is ready to receive a message and this therefore avoids a handshake between sender and receiver, increasing the performance of the implementation. If the receiver is not ready, the outcome of the send operation is unpredictable in terms of the buffer being ready and the MPI documentation deems the outcome undefined. It may be for example that values will be over-written before being consumed.
You might also like to view...
A malicious third party could try to decrypt a message using every possible decryption key, a process known as_______.
a) cracking. b) brute-force cracking. c) step-by-step cracking. d) None of the above.
If there is an if-else block within another if-else block, then the block is said to be __________.
A. compound B. nested C. cascading D. you can’t write if-else blocks within an if-else block.
To move the active cell up one row, press ____.
A. Shift + Enter B. Ctrl + Tab C. Shift + Tab D. Ctrl + Enter
How does the object-oriented view of component-level design differ from the traditional view?
What will be an ideal response?