Fatal Error in MPI_Irecv: Aborting Job

Question

The following sequence of errors is received when I try to run a problem on four processors. The MPI command I use is mpirun -np 4

I apologize for posting the error message as is (Primarily due a lack of knowledge on deciphering the information given). Would appreciate your input on the following:

What does the error message mean? At what point does one receive it? Is it because of the system memory (hardware) or is it due to a communication error (something related to MPI_Isend/Irecv?, i.e. Software issue).
Finally, how do I fix this?

Thanks!

ERROR message received follows below:: - - *PLEASE NOTE: This error is received only when the time is large*. Code computes fine when time required to compute data is small (i.e, 300 time steps compared to 1000 time steps)

aborting job:

Fatal error in MPI_Irecv: Other MPI error, error stack:

MPI_Irecv(143): MPI_Irecv(buf=0x8294a60, count=48, MPI_DOUBLE, src=2, tag=-1, MPI_COMM_WORLD, request=0xffffd68c) failed

MPID_Irecv(64): Out of memory

aborting job:

Fatal error in MPI_Irecv: Other MPI error, error stack:

MPI_Irecv(143): MPI_Irecv(buf=0x8295080, count=48, MPI_DOUBLE, src=3, tag=-1, MPI_COMM_WORLD, request=0xffffd690) failed

MPID_Irecv(64): Out of memory

aborting job: Fatal error in MPI_Isend: Internal MPI error!, error stack:

MPI_Isend(142): MPI_Isend(buf=0x8295208, count=48, MPI_DOUBLE, dest=3, tag=0, MPI_COMM_WORLD, request=0xffffd678) failed

(unknown)(): Internal MPI error!

aborting job: Fatal error in MPI_Irecv: Other MPI error, error stack:

MPI_Irecv(143): MPI_Irecv(buf=0x82959b0, count=48, MPI_DOUBLE, src=2, tag=-1, MPI_COMM_WORLD, request=0xffffd678) failed

MPID_Irecv(64): Out of memory

rank 3 in job 1 myocyte80_37021 caused collective abort of all ranks exit status of rank 3: return code 13

rank 1 in job 1 myocyte80_37021 caused collective abort of all ranks exit status of rank 1: return code 13

EDIT: (SOURCE CODE)

Header files
Variable declaration
TOTAL TIME = 
...
...
double *A = new double[Rows];
double *AA = new double[Rows];
double *B = new double[Rows;
double *BB = new double[Rows];
....
....
int Rmpi;
int my_rank;
int p;
int source; 
int dest;
int tag = 0;
function declaration

int main (int argc, char *argv[])
{
MPI_Status status[8]; 
MPI_Request request[8];
MPI_Init (&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &p);   
MPI_Comm_rank(MPI_COMM_WORLD, &my_rank);

//PROBLEM SPECIFIC PROPERTIES. VARY BASED ON NODE 
if (Flag = 1)
{
if (my_rank == 0)
{
Defining boundary (start/stop) for special elements in tissue (Rows x Column)
}
if (my_rank == 2)
..
if (my_rank == 3)
..
if (my_rank == 4)
..
}

//INITIAL CONDITIONS ALSO VARY BASED ON NODE
for (Columns = 0; Columns<48; i++) // Normal Direction
{
for (Rows = 0; Rows<48; y++)  //Transverse Direction
{
if (Flag =1 )
{
if (my_rank == 0)
{
Initial conditions for elements
}
if (my_rank == 1) //MPI
{
}
..
..
..
//SIMULATION START

while(t[0][0] < TOTAL TIME)
{       
for (Columns=0; Columns ++) //Normal Direction
{
for (Rows=0; Rows++) //Transverse Direction
{
//SOME MORE PROPERTIES BASED ON NODE
if (my_rank == 0)
{
if (FLAG = 1)
{
Condition 1
}   
 else
{
Condition 2 
}
}

if (my_rank = 1)
....
 ....
  ...

//Evaluate functions (differential equations)
Function 1 ();
Function 2 ();
...
...

//Based on output of differential equations, different nodes estimate variable values. Since   
 the problem is of nearest neighbor, corners and edges have different neighbors/ boundary   
 conditions
if (my_rank == 0)
{
If (Row/Column at bottom_left)                  
{
Variables =
}

if (Row/Column at Bottom Right) 
{
Variables =
}
}
...
 ...

 //Keeping track of time for each element in Row and Column. Time is updated for a certain  
 element. 
 t[Column][Row] = t[Column][Row]+dt;

  }
  }//END OF ROWS AND COLUMNS

 // MPI IMPLEMENTATION. AT END OF EVERY TIME STEP, Nodes communicate with nearest neighbor
 //First step is to populate arrays with values estimated above
 for (Columns, ++) 
 {
 for (Rows, ++) 
 {
 if (my_rank == 0)
 {
 //Loading the Edges of the (Row x Column) to variables. This One dimensional Array data 
 is shared with its nearest neighbor for computation at next time step.

 if (Column == 47)
 {
 A[i] = V[Column][Row]; 
 …
 }
 if (Row == 47)
 {
 B[i] = V[Column][Row]; 
 }
 }

...
...                 

 //NON BLOCKING MPI SEND RECV TO SHARE DATA WITH NEAREST NEIGHBOR

 if ((my_rank) == 0)
 {
 MPI_Isend(A, Rows, MPI_DOUBLE, my_rank+1, 0, MPI_COMM_WORLD, &request[1]);
 MPI_Irecv(AA, Rows, MPI_DOUBLE, my_rank+1, MPI_ANY_TAG, MPI_COMM_WORLD, &request[3]);
 MPI_Wait(&request[3], &status[3]);  
 MPI_Isend(B, Rows, MPI_DOUBLE, my_rank+2, 0, MPI_COMM_WORLD, &request[5]);
 MPI_Irecv(BB, Rows, MPI_DOUBLE, my_rank+2, MPI_ANY_TAG, MPI_COMM_WORLD, &request[7]);
 MPI_Wait(&request[7], &status[7]);
 }

if ((my_rank) == 1)
{
MPI_Irecv(CC, Rows, MPI_DOUBLE, my_rank-1, MPI_ANY_TAG, MPI_COMM_WORLD, &request[1]);
MPI_Wait(&request[1], &status[1]); 
MPI_Isend(Cmpi, Rows, MPI_DOUBLE, my_rank-1, 0, MPI_COMM_WORLD, &request[3]);

MPI_Isend(D, Rows, MPI_DOUBLE, my_rank+2, 0, MPI_COMM_WORLD, &request[6]); 
MPI_Irecv(DD, Rows, MPI_DOUBLE, my_rank+2, MPI_ANY_TAG, MPI_COMM_WORLD, &request[8]);
MPI_Wait(&request[8], &status[8]);
}

if ((my_rank) == 2)
{
MPI_Isend(E, Rows, MPI_DOUBLE, my_rank+1, 0, MPI_COMM_WORLD, &request[2]);
MPI_Irecv(EE, Rows, MPI_DOUBLE, my_rank+1, MPI_ANY_TAG, MPI_COMM_WORLD, &request[4]);
MPI_Wait(&request[4], &status[4]);

MPI_Irecv(FF, Rows, MPI_DOUBLE, my_rank-2, MPI_ANY_TAG, MPI_COMM_WORLD, &request[5]);
MPI_Wait(&request[5], &status[5]);
MPI_Isend(Fmpi, Rows, MPI_DOUBLE, my_rank-2, 0, MPI_COMM_WORLD, &request[7]);
}

if ((my_rank) == 3)
{
MPI_Irecv(GG, Rows, MPI_DOUBLE, my_rank-1, MPI_ANY_TAG, MPI_COMM_WORLD, &request[2]);
MPI_Wait(&request[2], &status[2]);
MPI_Isend(G, Rows, MPI_DOUBLE, my_rank-1, 0, MPI_COMM_WORLD, &request[4]);

MPI_Irecv(HH, Rows, MPI_DOUBLE, my_rank-2, MPI_ANY_TAG, MPI_COMM_WORLD, &request[6]);
MPI_Wait(&request[6], &status[6]); 
MPI_Isend(H, Rows, MPI_DOUBLE, my_rank-2, 0, MPI_COMM_WORLD, &request[8]);
}

 //RELOADING Data (from MPI_IRecv array to array used to compute at next time step)
 for (Columns, ++) 
 {
 for (Rows, ++) 
 {
 if (my_rank == 0)
 {
 if (Column == 47)
 {
 V[Column][Row]= A[i];
 }
 if (Row == 47)
 {
 V[Column][Row]=B[i];
 }
  }

  ….
 //PRINT TO OUTPUT FILE AT CERTAIN POINT
 printval = 100; 
 if ((printdata>=printval))
 {
 prttofile ();
 printdata = 0;
 }
 printdata = printdata+1;
 compute_dt (); 

 }//CLOSE ALL TIME STEPS

 MPI_Finalize ();

  }//CLOSE MAIN

"Out of memory" seems as clear as it gets to me. What are you asking exactly? — ildjarn
No one can accurately tell you why you're getting this error with this information alone. This question should be clarified to better describe the problem perhaps with a code snippet. As it is this will probably be closed as not a real question. — AJG85
Have you computed how much memory your problem requires, or just monitored it with top? Like @ildjam said, out of memory is pretty clear.. Especially when you say it happens at larger problem sizes, it's possible you've got a memory leak at each time step that simply compounds until you run out. — Adam
Adam: This is running on a cluster. I use the following command to submit a job: mpirun -np 4 <jobname>. I assumed it was running on 4 nodes. At our set up each node has 4 processors. However, when I use "top" I notice that, each of the 4 processors is running at 99% efficiency. The questions are the following: 1. What do you mean by memory leak? How do I go about this problem if it is a memory issue (ie. hardware based due to insufficient memory)? Basically, can I setup MPI in some way to fix this? 2. The way I submit an MPI RUN, how do i submit so that each node computes only 1 thread? — Ashmohan

David David · Accepted Answer · 2011-06-02T23:28:27

Are you repeatedly calling MPI_Irecv? If so, you may not realize that each call allocates a request handle - and these are freed when the message is received and tested for completion with (eg.) MPI_Test. It's possible you could exhaust memory with over-use of MPI_Irecv - or the memory assigned by an MPI implementation for this purpose.

Only seeing the code would confirm the problem.

Fatal Error in MPI_Irecv: Aborting Job

2 Answers