gaurav gupta wrote:
> Hello,
>
> I want to know that which task is running on which node. Is there any
> way to know this.
> Is there any profiling tool provided along with openmpi to calculate
> time taken in various steps.
>
> --
> GAURAV GUPTA
> B.Tech III Yr. , Department of Computer Science & Engineering
> IT BHU , Varanasi
> Contacts
> Phone No: +91-99569-49491
>
> e-mail :
> gaurav.gupta_at_[hidden] <mailto:gaurav.gupta_at_[hidden]>
> gaurav.gupta.cse06_at_[hidden] <mailto:gaurav.gupta.cse06_at_[hidden]>
> 1989.gaurav_at_[hidden] <mailto:1989.gaurav_at_[hidden]>
> ------------------------------------------------------------------------
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users
Hi Gupta,
I ran into the same problem. In my case I wanted to define the root node
on a specific host for a synchronization step using rsync between the
hosts running the processes. Here is some linux C code that might help
you. It builds an array mpi_host with the hostname of each node, and an
index array mpi_host_rank that shows which processes are running on the
same node. The BUG, MY_MALLOC and my_printf macro's are wrappers for
C-functions assert, malloc and printf. The code assumes name-resolution
is the same on all nodes.
#define LINE_MAX 1024
#define MPI_NPROCS_MAX 256
#define INVALID (-1)
int mpi_nprocs;
int mpi_id;
int mpi_nhosts;
int mpi_root_id;
char *mpi_hosts;
char *mpi_host[MPI_NPROCS_MAX];
int mpi_host_rank[MPI_NPROCS_MAX];
int main(void)
{
int iproc;
char hostname[LINE_MAX];
mpi_nprocs = 1;
mpi_id = 0;
mpi_nhosts = 1;
mpi_root_id = 0;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_nprocs);
BUG(mpi_nprocs > MPI_NPROCS_MAX)
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_id);
BUG(gethostname(hostname, LINE_MAX) != 0)
REGISTER_MALLOC(mpi_hosts, char, LINE_MAX * mpi_nprocs)
for (iproc = 0; iproc < mpi_nprocs; iproc++)
mpi_host[iproc] = mpi_hosts + iproc * LINE_MAX;
if (mpi_nprocs == 1)
strcpy(mpi_host[0], hostname);
else
MPI_Allgather(hostname, LINE_MAX, MPI_CHAR,
mpi_hosts, LINE_MAX, MPI_CHAR, MPI_COMM_WORLD);
MPI_Barrier(MPI_COMM_WORLD);
for (iproc = 0; iproc < mpi_nprocs; iproc++)
mpi_host_rank[iproc] = INVALID;
mpi_nhosts = 0;
for (iproc = 0; iproc < mpi_nprocs; iproc++)
{
int jproc;
if (mpi_host_rank[iproc] != INVALID) continue;
++mpi_nhosts;
BUG(mpi_nhosts > mpi_nprocs)
mpi_host_rank[iproc] = mpi_nhosts - 1;
for (jproc = iproc + 1; jproc < mpi_nprocs; jproc++)
{
if (mpi_host_rank[jproc] != INVALID) continue;
if (strcasecmp(mpi_host[jproc], mpi_host[iproc]) == 0)
mpi_host_rank[jproc] = mpi_host_rank[iproc];
}
}
//find specific host if available
mpi_root_id = 0;
for (iproc = 0; iproc < mpi_nprocs; iproc++)
{
if (strcasecmp(mpi_host[iproc], "nodep140") == 0)
{
mpi_root_id = iproc;
break;
}
}
BUG(mpi_nprocs < 1)
BUG(mpi_nhosts < 1)
my_printf("hostname=%s\n", hostname);
my_printf("mpi_nprocs=%d\n", mpi_nprocs);
my_printf("mpi_id=%d\n", mpi_id);
for (iproc = 0; iproc < mpi_nprocs; iproc++)
my_printf("iproc=%d host=%s\n", iproc, mpi_host[iproc]);
my_printf("mpi_nhosts=%d\n", mpi_nhosts);
for (iproc = 0; iproc < mpi_nprocs; iproc++)
my_printf("iproc=%d host_rank=%d\n", iproc, mpi_host_rank[iproc]);
my_printf("mpi_root_id=%d host=%s host rank=%d\n",
mpi_root_id, mpi_host[mpi_root_id], mpi_host_rank[mpi_root_id]);
}
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}