Message Passing Interface

Question 1

Purpose of MPI_Init

Answer 1

Initializes MPI

Question 2

Purpose of MPI_Finalize

Answer 2

Terminates MPI

Question 3

What do MPI functions return?

Answer 3

MPI_SUCCESS

Question 4

What is a communicator?

Answer 4

Abstract description of a group of processes.

Question 5

What is MPI_COMM_WORLD?

Answer 5

Describes all the processes involved in your parallel run.

Question 6

What is a processor rank?
How does one find the rank of a process within a communicator?

Answer 6

A unique identifier assigned to each process.

MPI_Comm_rank()

Question 7

How are ranks numbered within a communicator?

Answer 7

From zero to the number of processes minus 1

Question 8

What does MPI_Comm_Split do?

Answer 8

Divides the communicator into disjoint sub communications

Question 9

What role does the color argument to MPI_Comm_Split perform?

Answer 9

Controls of subset assignment.

Question 10

What role does the key argument to MPI_Comm_Split perform?

Answer 10

Control the ranking of the divided processes.

Question 11

What is global numbering?

Answer 11

Numbering the processes in the MPI World (all communicators).

Question 12

What is local numbering

Answer 12

Numbering the processes within a particular communicator

Question 13

What does it mean when an MPI program is loosely synchronous?

Answer 13

Tasks synchronize to perform interactions, otherwise, run asynchronously.

Question 14

What 3 pieces of information are used to pass sending data to a collective communication operation?

Answer 14

Send Buffer
Count
Data Type

Question 15

What function does MPI_barrier perform?

Answer 15

Blocks all processes until they all reach the barrier wall, synchronizing the processes.

Question 16

What function does MPI_Bcast perform?

Answer 16

Communicates data from one process to other processes.

Question 17

What function does MPI_Reduce perform?

Answer 17

Bring data from all processes into one single item.

Question 18

What is the difference between MPI_Reduce and MPI_Allreduce?

Answer 18

Allreduce returns the single result back to all the processes.

Question 19

What does MPI_Scatter do?

Answer 19

Each process receives a different subset of data.

Question 20

What does MPI_Gather do?

Answer 20

Collect data from all processes to a single root process.

Question 21

What does MPI_Alltoall do?

Answer 21

A collection of simultaneous broadcasts and gathers.

Question 22

What does MPI_Scan do?

Answer 22

Performs a running reduction, but it keeps the partial results.

Question 23

What does it mean if message-passing operations are buffered?

Answer 23

A send operation can complete no matter whether the receive has been posted.

Question 24

What does it mean if message-passing operations are blocking?

Answer 24

The sending process is blocked until the receiving process has received the message.

Question 25

How many calls does it take to receive a message using a non-blocking protocol?

Answer 25

2 Calls:

The first call initiates the receive operation and specifies the buffer size where the message will be stored.
The second call checks whether the receive operation has been completed.

Question 26

What impact does buffering have on program portability?

Answer 26

Different computers have different buffer sizes, so not all buffers are program portable.

Question 27

What techniques can be used to avoid deadlock in MPI communication?

Answer 27

Use nonblocking communication
Use tie-breaking to coordinate communication
Use MPI_Sendrecv call to break dependencies between send and recv calls.