Increasing the concurrency/parallelism for a nested model

[jstewa-snl]

I'm trying to run a nested model that basically does sampling within sampling and return the mean and standard deviation of the inner_method responses to the outer_method. Below is the dakota input that I currently have for this setup. It currently runs ok and returns the expected mean and standard deviation. However, with this setup I can only get 1 outer sample to run at a time with the 5 inner samples running concurrently. What I'd like to do is increase the evaluation throughput and run, for example, 10 simulations at the same time, i.e., 2 samples from the outer_method each with the 5 inner samples. After increasing the computing resources, I've tried setting evaluation_concurrency = 10 (doesn't make sense, but I tried it) and not setting an evaluation concurrency and let dakota automatically determine things (it only runs 1 job at a time). I've been running dakota 6.21 in serial since the actual simulations being performed are parallel simulations.

Any ideas on how to increase the concurrency/parallelism with this setup, if possible?

----------------------------------------------------------------------------

Top-level settings

environment
tabular_data
tabular_data_file = "result.txt"
top_method_pointer = "outer_method"

----------------------------------------------------------------------------

Method specification and settings

method
id_method = "outer_method"
output
verbose
sampling
samples = 10
seed = 13579
sample_type
lhs

refinement_samples = 10

model_pointer = "outer_model"

method
id_method = "inner_method"
output
verbose
sampling
sample_type
lhs
samples = 5
model_pointer = "inner_model"

----------------------------------------------------------------------------

Mapping of variables to responses through an interface

model
id_model = "outer_model"
nested
sub_method_pointer = "inner_method"
primary_variable_mapping = "a0"
primary_response_mapping = 1.0 0.0
0.0 1.0
variables_pointer = "outer_variables"
responses_pointer = "outer_responses"

model
id_model = "inner_model"
single
interface_pointer = "interface"
variables_pointer = "inner_variables"
responses_pointer = "inner_responses"

----------------------------------------------------------------------------

Parameter sets to be iterated by a particular method

variables
id_variables = "outer_variables"
active
design
mixed
continuous_design = 1
descriptors = "a"
lower_bounds = -1.0
upper_bounds = 1.0

variables
id_variables = "inner_variables"
active
design
mixed
continuous_state = 1
descriptors = "a0"
discrete_design_range = 1
descriptors = "x"
lower_bounds = 1
upper_bounds = 1000

----------------------------------------------------------------------------

How model function evaluations are performed

interface
id_interface = "interface"
analysis_drivers = "driver.inp"
fork
parameters_file = "params.inp"
results_file = "results.out"
parameters_format
aprepro
results_format
standard
file_save
work_directory
named = "WorkDirs/workdir"
directory_tag
directory_save
asynchronous
evaluation_concurrency = 5

----------------------------------------------------------------------------

Model output data returned by an interface

responses
id_responses = "outer_responses"
descriptors = "mu" "sigma"
response_functions = 2
no_gradients
no_hessians

responses
id_responses = "inner_responses"
descriptors = "energy"
response_functions = 1
no_gradients
no_hessians

----------------------------------------------------------------------------