A simple 2-qubit Bell circuit with 2000 shots took me 37 seconds of Qiskit Runtime usage on ibm_brisbane backend?
For the circuit of this size I would expect at least a several orders of magnitude smaller number. Where does the extra usage time come from and what kind of scaling should I expect for bigger circuits?
One of the sources of usage overhead that I expect is pulse scheduling and qubit initialization, which I would expect not to scale dramatically with the growing circuits. I also expect the backends with larger number of qubits to take longer to initialize. Let me know if this is correct.