University:
Boston University
Course:
MET CS 201 | Introduction to Programming
Academic year:

2020
Views:

296

Pages:

20
Author:

Mary Mueller

Blocking Receives and Sends  MPI_Recv is always blocking.  Blocking means the function call will not return until the receive is completed.  It is safe to use the received data right after calling MPI_Recv.  MPI_Send try not to block, but don’t guarantee it.  If the size of send data is smaller than that of the Send Queue, MPI_Send is not blocking --the data is sent to the Receive Queue without waiting.  But if the size of send data is larger than that of the Send Queue, MPI_Send is blocking --- it first sends a chunk of data, then stops sending when the Send Queue is full and will restart sending when the Send Queue becomes empty again (for example when the chunk of data has been moved to the Receive Queue).  The later case often happens, so it is OK to think that MPI_Send is blocking. A deadlock due to blocking receives  An example: swapping arrays between two processes.  The following code meets a deadlock situation and will hang forever.  Both processes are blocked at MPI_Recv no matter how large the data size is. int n=10; // a small data size int my_rank, n_send1[n], n_send2[n], n_recv1[n], n_recv2[n]; MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); if (my_rank==0){ MPI_Recv( &n_recv2, n, MPI_INT, 1, 11, MPI_COMM_WORLD, NULL); MPI_Send( &n_send1, n, MPI_INT, 1, 10, MPI_COMM_WORLD); } else if (my_rank==1){ MPI_Recv( &n_recv1, n, MPI_INT, 0, 10, MPI_COMM_WORLD, NULL); MPI_Send( &n_send2, n, MPI_INT, 0, 11, MPI_COMM_WORLD); } A deadlock due to blocking sends  If the sizes of the send arrays are large enough, MPI_Send becomes blocking, then the following code meets a deadlock situation.  Both processes are blocked at MPI_Send for a large data size. int n=5000; // a large data size int my_rank, n_send1[n], n_send2[n], n_recv1[n], n_recv2[n]; MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); if (my_rank==0){ MPI_Send( &n_send1, n, MPI_INT, 1, 10, MPI_COMM_WORLD); MPI_Recv( &n_recv2, n, MPI_INT, 1, 11, MPI_COMM_WORLD, NULL); } else if (my_rank==1){ MPI_Send( &n_send2, n, MPI_INT, 0, 11, MPI_COMM_WORLD); MPI_Recv( &n_recv1, n, MPI_INT, 0, 10, MPI_COMM_WORLD, NULL); } Break the deadlock (1)  Send and receive are coordinated, so there is no deadlock. int n=5000; // a large data size int my_rank, n_send1[n], n_send2[n], n_recv1[n], n_recv2[n]; MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); if (my_rank==0){ MPI_Send( &n_send1, n, MPI_INT, 1, 10, MPI_COMM_WORLD); MPI_Recv( &n_recv2, n, MPI_INT, 1, 11, MPI_COMM_WORLD, NULL); } else if (my_rank==1){ MPI_Recv( &n_recv1, n, MPI_INT, 0, 10, MPI_COMM_WORLD, NULL); MPI_Send( &n_send2, n, MPI_INT, 0, 11, MPI_COMM_WORLD); } Point-to-point communication (3): Sendrecv  The send-receive operations combine the sending of a message to one destination and the receiving of another message in one call.  MPI_Sendrecv executes a blocking send and receive operation.  Syntax: int MPI_Sendrecv (const void* senddata, int sendcount, MPI_Datatype sendtype, int dest, int sendtag, void* recvdata, int recvcount, MPI_Datatype recvtype, int source, int recvtag, MPI_Comm comm, MPI_Status* status)  The communicator for send and receive must be the same.  The destination for send and the source for receive may be the same or different.  The tags, count, datatype for send and receive may be the same or different.  The send buffer and the receive buffer must be disjoint. Break the deadlock (2)  Send and receive are automatically coordinated in MPI_Sendrecv, so there is no deadlock. int n=5000; // a large data size int my_rank, n_send1[n], n_send2[n], n_recv1[n], n_recv2[n]; MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); if (my_rank==0){ MPI_Sendrecv ( &n_send1, n, MPI_INT, 1, 10, &n_recv2, n, MPI_INT, 1, 11, MPI_COMM_WORLD, &status); } else if (my_rank==1){ MPI_Sendrecv ( &n_send2, n, MPI_INT, 0, 11, &n_recv1, n, MPI_INT, 0, 10, MPI_COMM_WORLD, &status); } Exercise 2: Circular shift and ring programs  Write two MPI codes (in C or Fortran) to do the following tasks respectively: 1) Circular shift program: Every process sends its rank to its right neighbor. The process with the largest rank send its rank to process 0. 2) Ring program: Pass a value around all processes in a ring-like fashion. The passing sequence is 0  1  …  N  0, where N is the maximum number of processes.  Hints: Use MPI_Send and MPI_Recv (or MPI_Sendrecv). Make sure every MPI_Send corresponds to a matching MPI_Recv.Be careful to avoid deadlocks. Synchronization: Barrier  Blocks until all processes in the communicator have reached this routine.  Syntax: int MPI_Barrier (MPI_Comm comm)  comm: Communicator. Print in order  Print in order using MPI_Barrier. int my_rank, my_size, i; MPI_Init(&argc, &argv); MPI_Comm_rank(MPI_COMM_WORLD, &my_rank); MPI_Comm_size(MPI_COMM_WORLD, &my_size); for(i=0; i