Discussion:
[OMPI users] fail to run client/server mode using MPI_COMM_ACCEPT/MPI_COMM_CONNECT without mpirun
王亮
2016-10-19 07:13:18 UTC
Permalink
I was doing the work to integrate OpenMPI into our application, and we did
not want to rely on mpirun.
And it seems that singleton MPI_INIT could meet our need. So I wrote a
client/sever demo to verify it.
However, it did not work. Is there anything wrong here or OpenMPI does not
support connect/accept work under singleton MPI_INIT?

The way I started client/server:
./server
./client $port_name(The port name is printed in server side)

I testes this on two versions(1.8.8 and 2.0.1)
On 1.8.8:
Trying connect to 2164981760.0;tcp://10.152.111.77,169.254.182.77:55937
+2164981761.0;tcp://10.152.111.77,169.254.182.77:21560:300
--------------------------------------------------------------------------
An attempt was made to send a message to a process whose
address is unknown:

Sender: [[50549,1],0]
From node: busaa024
Recipient: [[50549,1],0]
On node: busaa024
The message could not be delivered, and we are aborting.

On 2.0.1: timeout after 60s
Trying connect to 2540140922.0:2850261964
[busj50bda16.us.oracle.com:31952] OPAL ERROR: Timeout in file
base/pmix_base_fns.c at line 193
[busj50bda16:31952] *** An error occurred in MPI_Comm_connect
[busj50bda16:31952] *** reported by process [139681870123024,0]
[busj50bda16:31952] *** on communicator MPI_COMM_WORLD
[busj50bda16:31952] *** MPI_ERR_UNKNOWN: unknown error
[busj50bda16:31952] *** MPI_ERRORS_ARE_FATAL (processes in this
communicator will now abort,
[busj50bda16:31952] *** and potentially your MPI job)

The server code:
#include<mpi.h>
#include<stdio.h>
#include<stdlib.h>

int size, rank, msg;

int main(int argc, char *argv[]){
MPI_Comm cliente;
MPI_Status status;
char portname[MPI_MAX_PORT_NAME];

MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&size);
MPI_Comm_rank(MPI_COMM_WORLD,&rank);

MPI_Open_port(MPI_INFO_NULL, portname);
printf("portname: %s\n", portname);
MPI_Comm_accept(portname, MPI_INFO_NULL, 0, MPI_COMM_SELF, &cliente);
printf("client connected\n");
MPI_Recv(&msg, 1, MPI_INT, MPI_ANY_SOURCE, MPI_ANY_TAG, cliente,
&status);
printf("msg: %d\n", msg);
MPI_Comm_free(&cliente);
MPI_Close_port(portname);
MPI_Finalize();
}

The client code:
#include<mpi.h>
#include<stdio.h>
#include<stdlib.h>
#include<string.h>

int size, rank;
int main(int argc, char *argv[]){
MPI_Comm servidor;
int msg, tag, dest;
char portname[MPI_MAX_PORT_NAME];

MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD,&size);
MPI_Comm_rank(MPI_COMM_WORLD,&rank);
if (argc >= 2){
printf("Trying connect to %s\n", argv[1]);
strcpy(portname, argv[1]);
MPI_Comm_connect(portname, MPI_INFO_NULL, 0, MPI_COMM_WORLD,
&servidor);
msg = 42; tag = 0; dest = 0;
MPI_Send(&msg, 1, MPI_INT, dest, tag, servidor);
MPI_Comm_disconnect(&servidor);
}
MPI_Finalize();
}

Thanks,
Bright Wang

Loading...