The number of qubits supported by the MPI version of Qiskit Aer does not increase with the addition of nodes.

[Guogggg]

Informations

• Qiskit Aer version:0.16.0
• Python version:3.9.5
• Operating system:Ubuntu 20.04

What is the current behavior?

I have two compute nodes, each with 14GB of memory, and I have manually compiled the MPI version of Qiskit-Aer on both nodes. Now, I need to simulate a 30-qubit GHZ circuit, which requires around 16GB of RAM. In theory, my available RAM is 2 * 14GB = 28GB, which should be sufficient to run a 30-qubit circuit. My hostfile is set to node1:1, node2:1, and in the code, blocking_qubits = 27. I run the command mpirun -np 2 -machinefile hostfile python new_qiskit.py, but it throws an error indicating insufficient memory. When I checked with htop, I noticed that the memory usage on each node exceeds 14GB during computation. Here is Error message:

Abort(940225294) on node 1 (rank 1 in comm 0): Fatal error in PMPI_Wait: Unknown error class, error stack:
PMPI_Wait(206): MPI_Wait(request=0x7f127004bfc0, status=0x7f1278755050) failed
MPIR_Wait(106): 
(unknown)(): Unknown error class

Here is my code

from qiskit import QuantumCircuit
#from qiskit.providers.basic_provider import BasicSimulator
from qiskit.compiler import transpile, assemble
from qiskit_aer import AerSimulator

def create_ghz_circuit(n_qubits):
    circuit = QuantumCircuit(n_qubits)
    circuit.h(0)
    for qubit in range(n_qubits - 1):
        circuit.cx(qubit, qubit + 1)
    return circuit

shots = 1000
depth=10
qubits = 30
block_bits = 27
#backend = BasicSimulator()
backend = AerSimulator(method="statevector")

circuit = transpile(create_ghz_circuit(qubits),
                    optimization_level=0)
circuit.measure_all()

#result = backend.run(circuit, shots=shots, seed_simulator=10).result()
result = backend.run(circuit,shots=shots,seed_simulator=12345,blocking_enable=True,blocking_qubits=block_bits).result()
print(result)

adict = result.to_dict()
meta = adict['metadata']
print("meta: ", meta)
myrank = meta['mpi_rank']
print("myrank: ",myrank)

Steps to reproduce the problem

Running code with mpirun -np 2 -machinefile hostfile python new_qiskit.py

What is the expected behavior?

He should be up and running

Suggested solutions

[diehlpk]

I am running qiskit using MPI and according to the documentation (https://qiskit.github.io/qiskit-aer/howtos/running_gpu.html) you need to add blocking_enable=True, blocking_qubits=23 to run it distributed.