def main():
args = parse_args()
metrics.reset()
start_time = time.time()
logging.basicConfig(level=logging.INFO)
# Create a SimPy environment
env = simpy.Environment()
# Seed the random generator
random.seed(args.seed)
numpy.random.seed(args.seed)
# Parse network and get NetworkX object and ingress network list
network, ing_nodes = reader.read_network(args.network,
node_cap=10,
link_cap=10)
# Getting current SFC list, and the SF list of each SFC, and config
# use dummy placement and schedule for running simulator without algorithm
# TODO: make configurable via CLI
sf_placement = dummy_data.triangle_placement
schedule = dummy_data.triangle_schedule
# Getting current SFC list, and the SF list of each SFC, and config
sfc_list = reader.get_sfc(args.sf)
sf_list = reader.get_sf(args.sf, args.sfr)
config = reader.get_config(args.config)
# Create the simulator parameters object with the provided args
params = SimulatorParams(network,
ing_nodes,
sfc_list,
sf_list,
config,
args.seed,
sf_placement=sf_placement,
schedule=schedule)
log.info(params)
if args.trace:
trace = reader.get_trace(args.trace)
TraceProcessor(params, env, trace)
# Create a FlowSimulator object, pass the SimPy environment and params objects
simulator = FlowSimulator(env, params)
# Start the simulation
simulator.start()
# Run the simpy environment for the specified duration
env.run(until=args.duration)
# Record endtime and running_time metrics
end_time = time.time()
metrics.running_time(start_time, end_time)
# dump all metrics
log.info(metrics.metrics)
def setUp(self):
"""
Setup test environment
"""
logging.basicConfig(level=logging.ERROR)
self.env = simpy.Environment()
# Configure simulator parameters
network, ing_nodes, eg_nodes = reader.read_network(NETWORK_FILE,
node_cap=10,
link_cap=10)
sfc_list = reader.get_sfc(SERVICE_FUNCTIONS_FILE)
sf_list = reader.get_sf(SERVICE_FUNCTIONS_FILE, RESOURCE_FUNCTION_PATH)
config = reader.get_config(CONFIG_FILE)
self.metrics = Metrics(network, sf_list)
sf_placement = dummy_data.triangle_placement
schedule = dummy_data.triangle_schedule
# Initialize Simulator and SimulatoParams objects
self.simulator_params = SimulatorParams(log,
network,
ing_nodes,
eg_nodes,
sfc_list,
sf_list,
config,
self.metrics,
sf_placement=sf_placement,
schedule=schedule)
self.flow_simulator = FlowSimulator(self.env, self.simulator_params)
self.flow_simulator.start()
self.env.run(until=SIMULATION_DURATION)
class TestFlowSimulator(TestCase):
flow_simulator = None
simulator_params = None
def setUp(self):
"""
Setup test environment
"""
logging.basicConfig(level=logging.ERROR)
self.env = simpy.Environment()
# Configure simulator parameters
network, ing_nodes, eg_nodes = reader.read_network(NETWORK_FILE,
node_cap=10,
link_cap=10)
sfc_list = reader.get_sfc(SERVICE_FUNCTIONS_FILE)
sf_list = reader.get_sf(SERVICE_FUNCTIONS_FILE, RESOURCE_FUNCTION_PATH)
config = reader.get_config(CONFIG_FILE)
self.metrics = Metrics(network, sf_list)
sf_placement = dummy_data.triangle_placement
schedule = dummy_data.triangle_schedule
# Initialize Simulator and SimulatoParams objects
self.simulator_params = SimulatorParams(log,
network,
ing_nodes,
eg_nodes,
sfc_list,
sf_list,
config,
self.metrics,
sf_placement=sf_placement,
schedule=schedule)
self.flow_simulator = FlowSimulator(self.env, self.simulator_params)
self.flow_simulator.start()
self.env.run(until=SIMULATION_DURATION)
def test_simulator(self):
"""
Test the simulator
"""
# Collect metrics
self.metric_collection = self.metrics.get_metrics()
# Check if Simulator is initiated correctly
self.assertIsInstance(self.flow_simulator, FlowSimulator)
# Check if Params are set correctly
self.assertIsInstance(self.simulator_params, SimulatorParams)
# Check if generated flows are equal to processed flow + dropped + active flows
gen_flow_check = self.metric_collection['generated_flows'] == (
self.metric_collection['processed_flows'] +
self.metric_collection['dropped_flows'] +
self.metric_collection['total_active_flows'])
self.assertIs(gen_flow_check, True)
def init(self,
network_file,
service_functions_file,
config_file,
seed,
trace=None,
resource_functions_path=""):
# Initialize metrics, record start time
metrics.reset()
self.run_times = int(1)
self.start_time = time.time()
# Parse network and SFC + SF file
self.network, self.ing_nodes = reader.read_network(network_file,
node_cap=10,
link_cap=10)
self.sfc_list = reader.get_sfc(service_functions_file)
self.sf_list = reader.get_sf(service_functions_file,
resource_functions_path)
self.config = reader.get_config(config_file)
# Generate SimPy simulation environment
self.env = simpy.Environment()
# Instantiate the parameter object for the simulator.
self.params = SimulatorParams(self.network, self.ing_nodes,
self.sfc_list, self.sf_list, self.config,
seed)
# Trace handling
if trace:
trace = reader.get_trace(trace)
TraceProcessor(self.params, self.env, trace)
self.duration = self.params.run_duration
# Get and plant random seed
self.seed = seed
random.seed(self.seed)
numpy.random.seed(self.seed)
# Instantiate a simulator object, pass the environment and params
self.simulator = FlowSimulator(self.env, self.params)
# Start the simulator
self.simulator.start()
# Run the environment for one step to get initial stats.
self.env.step()
# Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# for changing node remaining capacities.
# Also, parse the network stats and prepare it in SimulatorState format.
self.parse_network()
self.network_metrics()
# Record end time and running time metrics
self.end_time = time.time()
metrics.running_time(self.start_time, self.end_time)
simulator_state = SimulatorState(self.network_dict,
self.simulator.params.sf_placement,
self.sfc_list, self.sf_list,
self.traffic, self.network_stats)
# self.writer.write_state_results(self.env, simulator_state)
return simulator_state
class Simulator(SimulatorInterface):
def __init__(self, test_mode=False, test_dir=None):
# Number of time the simulator has run. Necessary to correctly calculate env run time of apply function
self.run_times = int(1)
self.test_mode = test_mode
self.test_dir = test_dir
# Create CSV writer
self.writer = ResultWriter(self.test_mode, self.test_dir)
def init(self,
network_file,
service_functions_file,
config_file,
seed,
trace=None,
resource_functions_path=""):
# Initialize metrics, record start time
metrics.reset()
self.run_times = int(1)
self.start_time = time.time()
# Parse network and SFC + SF file
self.network, self.ing_nodes = reader.read_network(network_file,
node_cap=10,
link_cap=10)
self.sfc_list = reader.get_sfc(service_functions_file)
self.sf_list = reader.get_sf(service_functions_file,
resource_functions_path)
self.config = reader.get_config(config_file)
# Generate SimPy simulation environment
self.env = simpy.Environment()
# Instantiate the parameter object for the simulator.
self.params = SimulatorParams(self.network, self.ing_nodes,
self.sfc_list, self.sf_list, self.config,
seed)
# Trace handling
if trace:
trace = reader.get_trace(trace)
TraceProcessor(self.params, self.env, trace)
self.duration = self.params.run_duration
# Get and plant random seed
self.seed = seed
random.seed(self.seed)
numpy.random.seed(self.seed)
# Instantiate a simulator object, pass the environment and params
self.simulator = FlowSimulator(self.env, self.params)
# Start the simulator
self.simulator.start()
# Run the environment for one step to get initial stats.
self.env.step()
# Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# for changing node remaining capacities.
# Also, parse the network stats and prepare it in SimulatorState format.
self.parse_network()
self.network_metrics()
# Record end time and running time metrics
self.end_time = time.time()
metrics.running_time(self.start_time, self.end_time)
simulator_state = SimulatorState(self.network_dict,
self.simulator.params.sf_placement,
self.sfc_list, self.sf_list,
self.traffic, self.network_stats)
# self.writer.write_state_results(self.env, simulator_state)
return simulator_state
def apply(self, actions: SimulatorAction):
self.writer.write_action_result(self.env, actions)
# increase performance when debug logging is disabled
if logger.isEnabledFor(logging.DEBUG):
logger.debug(f"SimulatorAction: %s", repr(actions))
# Get the new placement from the action passed by the RL agent
# Modify and set the placement parameter of the instantiated simulator object.
self.simulator.params.sf_placement = actions.placement
# Update which sf is available at which node
for node_id, placed_sf_list in actions.placement.items():
available = {}
# Keep only SFs which still process
for sf, sf_data in self.simulator.params.network.nodes[node_id][
'available_sf'].items():
if sf_data['load'] != 0:
available[sf] = sf_data
# Add all SFs which are in the placement
for sf in placed_sf_list:
available[sf] = available.get(sf, {'load': 0.0})
self.simulator.params.network.nodes[node_id][
'available_sf'] = available
# Get the new schedule from the SimulatorAction
# Set it in the params of the instantiated simulator object.
self.simulator.params.schedule = actions.scheduling
# reset metrics for steps
metrics.reset_run()
# Run the simulation again with the new params for the set duration.
# Due to SimPy restraints, we multiply the duration by the run times because SimPy does not reset when run()
# stops and we must increase the value of "until=" to accomodate for this. e.g.: 1st run call runs for 100 time
# uniits (1 run time), 2nd run call will also run for 100 more time units but value of "until=" is now 200.
runtime_steps = self.duration * self.run_times
logger.debug("Running simulator until time step %s", runtime_steps)
self.env.run(until=runtime_steps)
# Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# for changing node remaining capacities.
# Also, parse the network stats and prepare it in SimulatorState format.
self.parse_network()
self.network_metrics()
# Increment the run times variable
self.run_times += 1
# Record end time of the apply round, doesn't change start time to show the running time of the entire
# simulation at the end of the simulation.
self.end_time = time.time()
metrics.running_time(self.start_time, self.end_time)
# Create a new SimulatorState object to pass to the RL Agent
simulator_state = SimulatorState(self.network_dict,
self.simulator.params.sf_placement,
self.sfc_list, self.sf_list,
self.traffic, self.network_stats)
self.writer.write_state_results(self.env, simulator_state)
return simulator_state
def parse_network(self) -> dict:
"""
Converts the NetworkX network in the simulator to a dict in a format specified in the SimulatorState class.
"""
self.network_dict = {'nodes': [], 'edges': []}
for node in self.params.network.nodes(data=True):
node_cap = node[1]['cap']
used_node_cap = node[1]['cap'] - node[1]['remaining_cap']
self.network_dict['nodes'].append({
'id': node[0],
'resource': node_cap,
'used_resources': used_node_cap
})
for edge in self.network.edges(data=True):
edge_src = edge[0]
edge_dest = edge[1]
edge_delay = edge[2]['delay']
edge_dr = edge[2]['cap']
# We use a fixed user data rate for the edges here as the functionality is not yet incorporated in the
# simulator.
# TODO: Implement used edge data rates in the simulator.
edge_used_dr = 0
self.network_dict['edges'].append({
'src': edge_src,
'dst': edge_dest,
'delay': edge_delay,
'data_rate': edge_dr,
'used_data_rate': edge_used_dr
})
def network_metrics(self):
"""
Processes the metrics and parses them in a format specified in the SimulatorState class.
"""
stats = metrics.get_metrics()
self.traffic = stats['run_total_requested_traffic']
self.network_stats = {
'total_flows': stats['generated_flows'],
'successful_flows': stats['processed_flows'],
'dropped_flows': stats['dropped_flows'],
'in_network_flows': stats['total_active_flows'],
'avg_end2end_delay': stats['avg_end2end_delay'],
'run_avg_end2end_delay': stats['run_avg_end2end_delay'],
'run_max_end2end_delay': stats['run_max_end2end_delay'],
'run_total_processed_traffic': stats['run_total_processed_traffic']
}
def init(self, seed):
# Reset predictor class at beginning of every init
if self.prediction:
self.predictor = TrafficPredictor(self.params, self.lstm_predictor)
# increment episode count
self.episode += 1
self.params.episode += 1
# reset network caps and available SFs:
reader.reset_cap(self.network)
# Initialize metrics, record start time
self.params.run_times = int(1)
self.start_time = time.time()
# Generate SimPy simulation environment
self.env = simpy.Environment()
self.env.process(self.writer.begin_writing(self.env, self.params))
self.params.metrics.reset_metrics()
# Instantiate the parameter object for the simulator.
if self.params.use_states and 'trace_path' in self.config:
logger.warning('Two state model and traces are both activated, thi will cause unexpected behaviour!')
if self.params.use_states:
if self.params.in_init_state:
self.params.in_init_state = False
# else:
self.env.process(self.params.start_mmpp(self.env))
self.duration = self.params.run_duration
# Get and plant random seed
self.seed = seed
random.seed(self.seed)
numpy.random.seed(self.seed)
self.params.reset_flow_lists()
# generate flow lists 1x here since we are in `init()`
self.params.generate_flow_lists()
# Instantiate a simulator object, pass the environment and params
self.simulator = FlowSimulator(self.env, self.params)
# Trace handling
if 'trace_path' in self.config:
TraceProcessor(self.params, self.env, self.trace, self.simulator)
# Start the simulator
self.simulator.start()
# TODO: Create runner here
controller_cls = eval(self.params.controller_class)
self.controller = controller_cls(self.env, self.params, self.simulator)
# # Run the environment for one step to get initial stats.
# self.env.step()
# # Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# # for changing node remaining capacities.
# # Also, parse the network stats and prepare it in SimulatorState format.
# self.parse_network()
# self.network_metrics()
# Record end time and running time metrics
self.end_time = time.time()
self.params.metrics.running_time(self.start_time, self.end_time)
simulator_state = self.controller.get_init_state()
# Check to see if traffic prediction is enabled to provide future traffic not current traffic
# if self.prediction:
# requested_traffic = self.get_current_ingress_traffic()
# self.predictor.predict_traffic(self.env.now, current_traffic=requested_traffic)
# stats = self.params.metrics.get_metrics()
# self.traffic = stats['run_total_requested_traffic']
# simulator_state = SimulatorState(self.network_dict, self.simulator.params.sf_placement, self.sfc_list,
# self.sf_list, self.traffic, self.network_stats)
# logger.debug(f"t={self.env.now}: {simulator_state}")
# set time stamp to calculate runtime of next apply call
self.last_apply_time = time.time()
# Check to see if init called in warmup, if so, set warmup to false
# This is to allow for better prediction and better overall control
# in the future
return simulator_state
class Simulator(SimulatorInterface):
def __init__(self, network_file, service_functions_file, config_file, resource_functions_path="", test_mode=False,
test_dir=None):
super().__init__(test_mode)
# Number of time the simulator has run. Necessary to correctly calculate env run time of apply function
self.network_file = network_file
self.test_dir = test_dir
# init network, sfc, sf, and config files
self.network, self.ing_nodes, self.eg_nodes = reader.read_network(self.network_file)
self.sfc_list = reader.get_sfc(service_functions_file)
self.sf_list = reader.get_sf(service_functions_file, resource_functions_path)
self.config = reader.get_config(config_file)
self.metrics = Metrics(self.network, self.sf_list)
# Assume result path is the path where network file is in.
self.result_base_path = os.path.dirname(self.network_file)
if 'trace_path' in self.config:
# Quick solution to copy trace file to same path for network file as provided by calling algo.
trace_path = os.path.join(os.getcwd(), self.config['trace_path'])
copyfile(trace_path, os.path.join(self.result_base_path, os.path.basename(trace_path)))
self.prediction = False
# Check if future ingress traffic setting is enabled
if 'future_traffic' in self.config and self.config['future_traffic']:
self.prediction = True
self.params = SimulatorParams(logger, self.network, self.ing_nodes, self.eg_nodes, self.sfc_list, self.sf_list,
self.config, self.metrics, prediction=self.prediction)
write_schedule = False
if 'write_schedule' in self.config and self.config['write_schedule']:
write_schedule = True
write_flow_actions = False
if 'write_flow_actions' in self.config and self.config['write_flow_actions']:
write_flow_actions = True
# Create CSV writer
self.writer = ResultWriter(self.test_mode, self.test_dir, write_schedule, write_flow_actions)
self.params.writer = self.writer
self.episode = 0
self.params.episode = 0
self.last_apply_time = None
# Load trace file
if 'trace_path' in self.config:
trace_path = os.path.join(os.getcwd(), self.config['trace_path'])
self.trace = reader.get_trace(trace_path)
self.lstm_predictor = None
if 'lstm_prediction' in self.config and self.config['lstm_prediction']:
self.lstm_predictor = LSTM_Predictor(self.trace, params=self.params,
weights_dir=self.config['lstm_weights'])
def __del__(self):
# write dropped flow locs to yaml
self.writer.write_dropped_flow_locs(self.metrics.metrics['dropped_flows_locs'])
def init(self, seed):
# Reset predictor class at beginning of every init
if self.prediction:
self.predictor = TrafficPredictor(self.params, self.lstm_predictor)
# increment episode count
self.episode += 1
self.params.episode += 1
# reset network caps and available SFs:
reader.reset_cap(self.network)
# Initialize metrics, record start time
self.params.run_times = int(1)
self.start_time = time.time()
# Generate SimPy simulation environment
self.env = simpy.Environment()
self.env.process(self.writer.begin_writing(self.env, self.params))
self.params.metrics.reset_metrics()
# Instantiate the parameter object for the simulator.
if self.params.use_states and 'trace_path' in self.config:
logger.warning('Two state model and traces are both activated, thi will cause unexpected behaviour!')
if self.params.use_states:
if self.params.in_init_state:
self.params.in_init_state = False
# else:
self.env.process(self.params.start_mmpp(self.env))
self.duration = self.params.run_duration
# Get and plant random seed
self.seed = seed
random.seed(self.seed)
numpy.random.seed(self.seed)
self.params.reset_flow_lists()
# generate flow lists 1x here since we are in `init()`
self.params.generate_flow_lists()
# Instantiate a simulator object, pass the environment and params
self.simulator = FlowSimulator(self.env, self.params)
# Trace handling
if 'trace_path' in self.config:
TraceProcessor(self.params, self.env, self.trace, self.simulator)
# Start the simulator
self.simulator.start()
# TODO: Create runner here
controller_cls = eval(self.params.controller_class)
self.controller = controller_cls(self.env, self.params, self.simulator)
# # Run the environment for one step to get initial stats.
# self.env.step()
# # Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# # for changing node remaining capacities.
# # Also, parse the network stats and prepare it in SimulatorState format.
# self.parse_network()
# self.network_metrics()
# Record end time and running time metrics
self.end_time = time.time()
self.params.metrics.running_time(self.start_time, self.end_time)
simulator_state = self.controller.get_init_state()
# Check to see if traffic prediction is enabled to provide future traffic not current traffic
# if self.prediction:
# requested_traffic = self.get_current_ingress_traffic()
# self.predictor.predict_traffic(self.env.now, current_traffic=requested_traffic)
# stats = self.params.metrics.get_metrics()
# self.traffic = stats['run_total_requested_traffic']
# simulator_state = SimulatorState(self.network_dict, self.simulator.params.sf_placement, self.sfc_list,
# self.sf_list, self.traffic, self.network_stats)
# logger.debug(f"t={self.env.now}: {simulator_state}")
# set time stamp to calculate runtime of next apply call
self.last_apply_time = time.time()
# Check to see if init called in warmup, if so, set warmup to false
# This is to allow for better prediction and better overall control
# in the future
return simulator_state
def apply(self, actions):
logger.debug(f"t={self.env.now}: {actions}")
# calc runtime since last apply (or init): that's the algorithm's runtime without simulation
alg_runtime = time.time() - self.last_apply_time
self.writer.write_runtime(alg_runtime)
simulator_state = self.controller.get_next_state(actions)
# # Get the new placement from the action passed by the RL agent
# # Modify and set the placement parameter of the instantiated simulator object.
# self.simulator.params.sf_placement = actions.placement
# Update which sf is available at which node
# for node_id, placed_sf_list in actions.placement.items():
# available = {}
# # Keep only SFs which still process
# for sf, sf_data in self.simulator.params.network.nodes[node_id]['available_sf'].items():
# if sf_data['load'] != 0:
# available[sf] = sf_data
# # Add all SFs which are in the placement
# for sf in placed_sf_list:
# if sf not in available.keys():
# available[sf] = available.get(sf, {
# 'load': 0.0,
# 'last_active': self.env.now,
# 'startup_time': self.env.now
# })
# self.simulator.params.network.nodes[node_id]['available_sf'] = available
# Get the new schedule from the SimulatorAction
# Set it in the params of the instantiated simulator object.
# self.simulator.params.schedule = actions.scheduling
# reset metrics for steps
self.params.metrics.reset_run_metrics()
# Run the simulation again with the new params for the set duration.
# Due to SimPy restraints, we multiply the duration by the run times because SimPy does not reset when run()
# stops and we must increase the value of "until=" to accomodate for this. e.g.: 1st run call runs for 100 time
# uniits (1 run time), 2nd run call will also run for 100 more time units but value of "until=" is now 200.
# runtime_steps = self.duration * self.params.run_times
# logger.debug("Running simulator until time step %s", runtime_steps)
# self.env.run(until=runtime_steps)
# Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# for changing node remaining capacities.
# Also, parse the network stats and prepare it in SimulatorState format.
# self.parse_network()
# self.network_metrics()
# Increment the run times variable
self.params.run_times += 1
# Record end time of the apply round, doesn't change start time to show the running time of the entire
# simulation at the end of the simulation.
self.end_time = time.time()
self.params.metrics.running_time(self.start_time, self.end_time)
# if self.params.use_states:
# self.params.update_state()
# generate flow data for next run (used for prediction)
self.params.generate_flow_lists(now=self.env.now)
# Check to see if traffic prediction is enabled to provide future traffic not current traffic
# if self.prediction:
# requested_traffic = self.get_current_ingress_traffic()
# self.predictor.predict_traffic(self.env.now, current_traffic=requested_traffic)
# stats = self.params.metrics.get_metrics()
# self.traffic = stats['run_total_requested_traffic']
# # Create a new SimulatorState object to pass to the RL Agent
# simulator_state = SimulatorState(self.network_dict, self.simulator.params.sf_placement, self.sfc_list,
# self.sf_list, self.traffic, self.network_stats)
# self.writer.write_state_results(self.episode, self.env.now, simulator_state,
# self.params.metrics.get_metrics())
logger.debug(f"t={self.env.now}: {simulator_state}")
# set time stamp to calculate runtime of next apply call
self.last_apply_time = time.time()
return simulator_state
def get_current_ingress_traffic(self) -> float:
"""
Get current ingress traffic for the LSTM module
Current limitation: works for 1 SFC and 1 ingress node
"""
# Get name of ingress SF from first SFC
first_sfc = list(self.sfc_list.keys())[0]
ingress_sf = self.params.sfc_list[first_sfc][0]
ingress_node = self.params.ing_nodes[0][0]
ingress_traffic = self.metrics.metrics['run_total_requested_traffic'][ingress_node][first_sfc][ingress_sf]
return ingress_traffic
def get_active_ingress_nodes(self):
"""Return names of all ingress nodes that are currently active, ie, produce flows."""
return [ing[0] for ing in self.ing_nodes if self.params.inter_arr_mean[ing[0]] is not None]
def init(self, seed):
# reset network caps and available SFs:
reader.reset_cap(self.network)
# Initialize metrics, record start time
metrics.reset_metrics()
self.run_times = int(1)
self.start_time = time.time()
# Parse network and SFC + SF file
# Generate SimPy simulation environment
self.env = simpy.Environment()
self.params = SimulatorParams(self.network, self.ing_nodes,
self.sfc_list, self.sf_list, self.config)
# Instantiate the parameter object for the simulator.
if self.params.use_states and 'trace_path' in self.config:
logger.warning(
'Two state model and traces are both activated, thi will cause unexpected behaviour!'
)
if self.params.use_states:
if self.params.in_init_state:
self.params.in_init_state = False
else:
self.params.update_state()
self.duration = self.params.run_duration
# Get and plant random seed
self.seed = seed
random.seed(self.seed)
numpy.random.seed(self.seed)
# Instantiate a simulator object, pass the environment and params
self.simulator = FlowSimulator(self.env, self.params)
# Start the simulator
self.simulator.start()
# Trace handling
if 'trace_path' in self.config:
trace_path = os.path.join(os.getcwd(), self.config['trace_path'])
trace = reader.get_trace(trace_path)
TraceProcessor(self.params, self.env, trace, self.simulator)
# Run the environment for one step to get initial stats.
self.env.step()
# Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# for changing node remaining capacities.
# Also, parse the network stats and prepare it in SimulatorState format.
self.parse_network()
self.network_metrics()
# Record end time and running time metrics
self.end_time = time.time()
metrics.running_time(self.start_time, self.end_time)
simulator_state = SimulatorState(self.network_dict,
self.simulator.params.sf_placement,
self.sfc_list, self.sf_list,
self.traffic, self.network_stats)
logger.debug(f"t={self.env.now}: {simulator_state}")
return simulator_state
class Simulator(SimulatorInterface):
def __init__(self,
network_file,
service_functions_file,
config_file,
resource_functions_path="",
test_mode=False,
test_dir=None):
super().__init__(test_mode)
# Number of time the simulator has run. Necessary to correctly calculate env run time of apply function
self.run_times = int(1)
self.network_file = network_file
self.test_dir = test_dir
# Create CSV writer
self.writer = ResultWriter(self.test_mode, self.test_dir)
# init network, sfc, sf, and config files
self.network, self.ing_nodes, self.eg_nodes = reader.read_network(
self.network_file)
self.sfc_list = reader.get_sfc(service_functions_file)
self.sf_list = reader.get_sf(service_functions_file,
resource_functions_path)
self.config = reader.get_config(config_file)
self.metrics = Metrics(self.network, self.sf_list)
# Assume result path is the path where network file is in.
self.result_base_path = os.path.dirname(self.network_file)
if 'trace_path' in self.config:
# Quick solution to copy trace file to same path for network file as provided by calling algo.
trace_path = os.path.join(os.getcwd(), self.config['trace_path'])
copyfile(
trace_path,
os.path.join(self.result_base_path,
os.path.basename(trace_path)))
self.prediction = False
# Check if future ingress traffic setting is enabled
if 'future_traffic' in self.config and self.config['future_traffic']:
self.prediction = True
self.params = SimulatorParams(self.network,
self.ing_nodes,
self.eg_nodes,
self.sfc_list,
self.sf_list,
self.config,
self.metrics,
prediction=self.prediction)
self.episode = 0
def __del__(self):
# write dropped flow locs to yaml
self.writer.write_dropped_flow_locs(
self.metrics.metrics['dropped_flows_locs'])
def init(self, seed):
# Reset predictor class at beginning of every init
if self.prediction:
self.predictor = TrafficPredictor(self.params)
# increment episode count
self.episode += 1
# reset network caps and available SFs:
reader.reset_cap(self.network)
# Initialize metrics, record start time
self.run_times = int(1)
self.start_time = time.time()
# Generate SimPy simulation environment
self.env = simpy.Environment()
self.params.metrics.reset_metrics()
# Instantiate the parameter object for the simulator.
if self.params.use_states and 'trace_path' in self.config:
logger.warning(
'Two state model and traces are both activated, thi will cause unexpected behaviour!'
)
if self.params.use_states:
if self.params.in_init_state:
self.params.in_init_state = False
else:
self.params.update_state()
self.duration = self.params.run_duration
# Get and plant random seed
self.seed = seed
random.seed(self.seed)
numpy.random.seed(self.seed)
self.params.reset_flow_lists()
# generate flow lists 1x here since we are in `init()`
self.params.generate_flow_lists()
# Instantiate a simulator object, pass the environment and params
self.simulator = FlowSimulator(self.env, self.params)
# Trace handling
if 'trace_path' in self.config:
trace_path = os.path.join(os.getcwd(), self.config['trace_path'])
trace = reader.get_trace(trace_path)
TraceProcessor(self.params, self.env, trace, self.simulator)
# Start the simulator
self.simulator.start()
# Run the environment for one step to get initial stats.
self.env.step()
# Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# for changing node remaining capacities.
# Also, parse the network stats and prepare it in SimulatorState format.
self.parse_network()
self.network_metrics()
# Record end time and running time metrics
self.end_time = time.time()
self.params.metrics.running_time(self.start_time, self.end_time)
# Check to see if traffic prediction is enabled to provide future traffic not current traffic
if self.prediction:
self.predictor.predict_traffic(self.env.now)
stats = self.params.metrics.get_metrics()
self.traffic = stats['run_total_requested_traffic']
simulator_state = SimulatorState(self.network_dict,
self.simulator.params.sf_placement,
self.sfc_list, self.sf_list,
self.traffic, self.network_stats)
logger.debug(f"t={self.env.now}: {simulator_state}")
# Check to see if init called in warmup, if so, set warmup to false
# This is to allow for better prediction and better overall control
# in the future
return simulator_state
def apply(self, actions: SimulatorAction):
self.writer.write_action_result(self.episode, self.env.now, actions)
logger.debug(f"t={self.env.now}: {actions}")
# Get the new placement from the action passed by the RL agent
# Modify and set the placement parameter of the instantiated simulator object.
self.simulator.params.sf_placement = actions.placement
# Update which sf is available at which node
for node_id, placed_sf_list in actions.placement.items():
available = {}
# Keep only SFs which still process
for sf, sf_data in self.simulator.params.network.nodes[node_id][
'available_sf'].items():
if sf_data['load'] != 0:
available[sf] = sf_data
# Add all SFs which are in the placement
for sf in placed_sf_list:
available[sf] = available.get(sf, {'load': 0.0})
self.simulator.params.network.nodes[node_id][
'available_sf'] = available
# Get the new schedule from the SimulatorAction
# Set it in the params of the instantiated simulator object.
self.simulator.params.schedule = actions.scheduling
# reset metrics for steps
self.params.metrics.reset_run_metrics()
# Run the simulation again with the new params for the set duration.
# Due to SimPy restraints, we multiply the duration by the run times because SimPy does not reset when run()
# stops and we must increase the value of "until=" to accomodate for this. e.g.: 1st run call runs for 100 time
# uniits (1 run time), 2nd run call will also run for 100 more time units but value of "until=" is now 200.
runtime_steps = self.duration * self.run_times
logger.debug("Running simulator until time step %s", runtime_steps)
self.env.run(until=runtime_steps)
# Parse the NetworkX object into a dict format specified in SimulatorState. This is done to account
# for changing node remaining capacities.
# Also, parse the network stats and prepare it in SimulatorState format.
self.parse_network()
self.network_metrics()
# Increment the run times variable
self.run_times += 1
# Record end time of the apply round, doesn't change start time to show the running time of the entire
# simulation at the end of the simulation.
self.end_time = time.time()
self.params.metrics.running_time(self.start_time, self.end_time)
if self.params.use_states:
self.params.update_state()
# generate flow data for next run (used for prediction)
self.params.generate_flow_lists(now=self.env.now)
# Check to see if traffic prediction is enabled to provide future traffic not current traffic
if self.prediction:
self.predictor.predict_traffic(self.env.now)
stats = self.params.metrics.get_metrics()
self.traffic = stats['run_total_requested_traffic']
# Create a new SimulatorState object to pass to the RL Agent
simulator_state = SimulatorState(self.network_dict,
self.simulator.params.sf_placement,
self.sfc_list, self.sf_list,
self.traffic, self.network_stats)
self.writer.write_state_results(self.episode, self.env.now,
simulator_state)
logger.debug(f"t={self.env.now}: {simulator_state}")
return simulator_state
def parse_network(self) -> dict:
"""
Converts the NetworkX network in the simulator to a dict in a format specified in the SimulatorState class.
"""
max_node_usage = self.params.metrics.get_metrics(
)['run_max_node_usage']
self.network_dict = {'nodes': [], 'edges': []}
for node in self.params.network.nodes(data=True):
node_cap = node[1]['cap']
run_max_node_usage = max_node_usage[node[0]]
# 'used_resources' here is the max usage for the run.
self.network_dict['nodes'].append({
'id':
node[0],
'resource':
node_cap,
'used_resources':
run_max_node_usage
})
for edge in self.network.edges(data=True):
edge_src = edge[0]
edge_dest = edge[1]
edge_delay = edge[2]['delay']
edge_dr = edge[2]['cap']
# We use a fixed user data rate for the edges here as the functionality is not yet incorporated in the
# simulator.
# TODO: Implement used edge data rates in the simulator.
edge_used_dr = 0
self.network_dict['edges'].append({
'src': edge_src,
'dst': edge_dest,
'delay': edge_delay,
'data_rate': edge_dr,
'used_data_rate': edge_used_dr
})
def network_metrics(self):
"""
Processes the metrics and parses them in a format specified in the SimulatorState class.
"""
stats = self.params.metrics.get_metrics()
self.traffic = stats['run_total_requested_traffic']
self.network_stats = {
'processed_traffic': stats['run_total_processed_traffic'],
'total_flows': stats['generated_flows'],
'successful_flows': stats['processed_flows'],
'dropped_flows': stats['dropped_flows'],
'in_network_flows': stats['total_active_flows'],
'avg_end2end_delay': stats['avg_end2end_delay'],
'run_avg_end2end_delay': stats['run_avg_end2end_delay'],
'run_max_end2end_delay': stats['run_max_end2end_delay'],
'run_avg_path_delay': stats['run_avg_path_delay'],
'run_total_processed_traffic':
stats['run_total_processed_traffic'],
'run_dropped_flows_per_node': stats['run_dropped_flows_per_node']
}
def get_active_ingress_nodes(self):
"""Return names of all ingress nodes that are currently active, ie, produce flows."""
return [
ing[0] for ing in self.ing_nodes
if self.params.inter_arr_mean[ing[0]] is not None
]
