On May 11, 2005, at 16:00, Jeff Squyres wrote:
> I assume you're using some flavor of BSEND, right?
Yes.
> Second: even if you do use BSEND, it should trigger at least some
> progress every time you invoke BSEND (or most any other communication
> MPI function).
No, MPI_Bsend does never trigger any progress on pending messages if each
individual message is larger than the short protocol! This leads to the
bizarre situation that a sender that is *exclusively* using MPI_Bsend will
never get a message through before he calls MPI_Finalize. Before that
he will die from buffer exhaustion, of course. Run the program
below and you will see...
/* Bsend.c - demonstrate the futility of buffered sends */
#include <stdio.h>
#include <getopt.h>
#include <stdlib.h>
#include <time.h>
#include <unistd.h>
#include <mpi.h>
int main (int argc, char *argv[]) {
int myrank, nprocs, i, k;
char *buf, *mpibuf, c;
MPI_Request req;
MPI_Status status;
time_t T;
long bufsize;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
bufsize = 100000; /* override with Bsend -b <bufsize> */
while ((c = getopt (argc, argv, "b:")) != -1)
{
switch (c)
{
case 'b':
bufsize = atol (optarg);
break;
}
}
buf = (char*)malloc (bufsize*sizeof(char));
MPI_Barrier (MPI_COMM_WORLD);
if (myrank == 0) {
printf ("Buffer size is %ld\n", bufsize*sizeof(char));
mpibuf = (char*)malloc (3*bufsize*sizeof(char));
/* let MPI buffer at least 3 messages */
MPI_Buffer_attach (mpibuf, 4*bufsize*sizeof(char));
for (k = 0; k < 10; k++) {
time(&T);
printf("sending message %d at %s", k+1, ctime(&T));
MPI_Bsend (buf, bufsize, MPI_CHAR, nprocs-1, 99, MPI_COMM_WORLD);
sleep (5);
}
MPI_Buffer_detach (mpibuf, &i);
free (mpibuf);
}
else if (myrank == nprocs-1) {
for (k = 0; k < 10; k++) {
MPI_Recv (buf, bufsize, MPI_CHAR, 0, 99, MPI_COMM_WORLD, &status);
time(&T);
printf ("****** message %d received at %s", k+1, ctime(&T));
}
}
free (buf);
MPI_Finalize();
return EXIT_SUCCESS;
}
--
Stephan Mertens @ http://www.uni-magdeburg.de/mertens
Supercomputing in Magdeburg @ http://tina.nat.uni-magdeburg.de
|
