Running SimPy on Multiple Processors with Parallel Python¶
Introduction¶
With SimPy 2.0, you can easily increase the performance of your simulation by using Parallel Python if you have a larger number of independent processors (multiple CPUs or cores). Parallel Python can distribute the execution of your SimPy processes to all cores of your CPU and even to other computers. You should read the PP documentation for further information on how this works.
Please, note that Parallel Python is not included in the SimPy distribution and needs to be downloaded <http://www.parallelpython.com/> and installed separately.
Examples¶
Example #1¶
The files PPExample.txt and PPExampleProcess.txt contain a small example with several car processes. It is important, that processes etc.are not defined in the file that starts the PP job server and executes the jobs, since ppserver.submit() only takes functions defined in the same file and module names to import, but no classes.
PPExample.py:
"""
Example for SimPy with Parallel Python.
"""
import PPExampleProcess
from SimPy.Simulation import *
import pp
def runSimulation(jobNum, numCars):
sim = SimPy.Simulation.Simulation()
cars = []
for i in range(numCars):
car = PPExampleProcess.Car(sim, i * jobNum + i)
sim.activate(car, car.run(), at = 0)
cars.append(car)
sim.simulate(until = 30)
server = pp.Server(ppservers = ())
for i in range(4):
job = server.submit(
runSimulation,
(i, 2),
(),
('SimPy.Simulation', 'PPExampleProcess'))
job()
PPExampleProcess.py:
from SimPy.Simulation import *
class Car(Process):
def __init__(self, sim, id):
Process.__init__(self, sim = sim)
self.id = id
def run(self):
while True:
yield hold, self, 10
print 'Car #%i at t = %i' % (self.id, self.sim.now())
The simulated process in this case is a simple car, that holds for ten steps and then prints the current simulation time. Obviously, each car process is independent from the other ones. Thus if we want to simulate a great number of cars, we can easily distribute the processes to many CPU cores and others computers in our network.
PP pickles everything it sends to other cores/computers. Since SimPy is currently not pickleable, you cannot submit Simulation.simulate() to the PPServer. In this example runSimulation is defined and submitted to the server. The code within it will be executed on each core/computer. In the example we create four simulation jobs with two cars for each job.
Run PPExample.py to execute the example.
Example #2¶
Files simulator.txt and processes.txt contain an example that simulates refrigerators in single-thread and in parallel simulation mode.
Class Simulator simulates a number of fridges and gets the resulting data.
Class ParallelSimulator simulates a number of fridges and gets the resulting data. A number of jobs will be created that use all available CPU cores or even other computers.
To use clustering, ParallelPython needs to be installed on all computers and the server demon “ppserver.py” must be started. The list of the servers’ IP addresses must then be passed to the constructor of this class.
Run simulator.py to execute this example.
File simulator.py:
# coding=utf8
"""
The fridge simulation
@author: Stefan Scherfke
@contact: stefan.scherfke at uni-oldenburg.de
"""
from time import clock
import logging
from SimPy.Simulation import Simulation, activate, initialize, simulate
import pp
from processes import Fridge, FridgeObserver
log = logging.getLogger('Simulator')
class Simulator(object):
"""
This class simulates a number of fridges and gets the resulting data.
"""
def __init__(self, numFridges, tau, aggSteps, duration):
"""
Setup the simulation with the specified number of fridges.
Tau specifies the simulation step for each frige. Furthermore the
observer will collect data each tau. Collected data
will be aggregated at the end of each aggSteps simulation steps.
@param numFridges: The number of simulated fridges
@type numFridges: unsigned int
@param tau: simulation step size for collecting data and simulating
the fridge
@type tau: float
@param aggSteps: Collected data will be aggregated each aggSteps
simulation steps. Signals interval will be
tau * aggSteps
@type aggSteps: unsigned int
@param duration: Duration of the simulation in hours
@type duration: unsigned int
"""
log.info('Initializing simulator ...')
self.simEnd = duration
self.sim = Simulation()
fridgeProperties = {'tau': tau}
self._fridges = []
for i in range(numFridges):
fridge = Fridge(self.sim, **fridgeProperties)
self._fridges.append(fridge)
self._observer = FridgeObserver(self.sim, self._fridges, tau, aggSteps)
def simulate(self):
"""
Initialize the system, start the simulation and return the collected
data.
@return: The fridgerators consumption after each aggregation
"""
log.info('Running simulation ...')
self.sim.initialize()
for fridge in self._fridges:
self.sim.activate(fridge, fridge.run(), at = 0)
self.sim.activate(self._observer, self._observer.run(), at = 0)
self.sim.simulate(until = self.simEnd)
log.info('Simulation run finished.')
return self._observer.getData()
class ParallelSimulator(object):
"""
This class simulates a number of fridges and gets the resulting data.
Unlike simulator, a number of jobs will be created that use all availale
CPU cores or even other computers.
To use clustering, ParallelPython needs to be installed on all computers
and the server demon "ppserver.py" must be started. The list of the server's
IPs must then be passed to the constructor of this class.
"""
def __init__(self, numFridges, tau, aggSteps, duration,
jobSize = 100, servers = ()):
"""
Setup the simulation with the specified number of fridges. It will be
split up in several parallel jobs, each with the specified number of
jobs.
Tau specifies the simulation step for each frige. Furthermore the
observer will collect data each tau. Collected data
will be aggregated at the end of each aggSteps simulation steps.
@param numFridges: The number of simulated fridges
@type numFridges: unsigned int
@param tau: simulation step size for collecting data and simulating
the fridge
@type tau: float
@param aggSteps: Collected data will be aggregated each aggSteps
simulation steps. Signals interval will be
tau * aggSteps
@type aggSteps: unsigned int
@param duration: Duration of the simulation
@type duration: unsigned int
@param jobSize: The number of friges per job, defaults to 100.
@type jobSize: unsigned int
@param servers: A list of IPs from on which the simulation shall be
executed. Defaults to "()" (use only SMP)
@type servers: tuple of string
"""
log.info('Initializing prallel simulation ...')
self._jobSize = jobSize
self._servers = servers
self._numFridges = numFridges
self._tau = tau
self._aggSteps = aggSteps
self.simEnd = duration
def simulate(self):
"""
Create some simulation jobs, run them and retrieve their results.
@return: The fridgerators consumption after each aggregation
"""
log.info('Running parallel simulation ...')
oldLevel = log.getEffectiveLevel() # pp changes the log level :(
jobServer = pp.Server(ppservers = self._servers)
# Start the jobs
remainingFridges = self._numFridges
jobs = []
while remainingFridges > 0:
jobs.append(jobServer.submit(self.runSimulation,
(min(self._jobSize, remainingFridges),),
(),
("logging", "SimPy.Simulation", "processes")))
remainingFridges -= self._jobSize
log.info('Number of jobs for simulation: %d' % len(jobs))
# Add each job's data
pSum = [0] * int((60 / self._aggSteps) * self.simEnd)
for job in jobs:
data = job()
for i in range(len(data)):
pSum[i] += data[i]
for s in pSum:
s /= len(jobs)
log.setLevel(oldLevel)
log.info('Parallel simulation finished.')
return pSum
def runSimulation(self, numFridges):
"""
Create a job with the specified number of fridges and controllers and
one observer. Simulate this and return the results.
@param numFridges: The number of fridges to use for this job
@type numFridges: unsigned int
@return: A list with the aggregated fridge consumption
"""
sim = SimPy.Simulation.Simulation()
sim.initialize()
fridgeProperties = {'tau': self._tau}
fridges = []
for i in range(numFridges):
fridge = processes.Fridge(sim, **fridgeProperties)
fridges.append(fridge)
sim.activate(fridge, fridge.run(), at = 0)
observer = processes.FridgeObserver(sim,
fridges, self._tau, self._aggSteps)
sim.activate(observer, observer.run(), at = 0)
sim.simulate(until = self.simEnd)
return observer.getData()
if __name__ == '__main__':
logging.basicConfig(
level = logging.INFO,
format = '%(asctime)s %(levelname)8s: %(name)s: %(message)s')
numFridges = 5000
tau = 1./60
aggStep = 15
duration = 4 + tau
sim = Simulator(numFridges, tau, aggStep, duration)
data = sim.simulate()
log.info('Results: ' + str(data))
servers = ()
sim = ParallelSimulator(numFridges, tau, aggStep, duration, 100, servers)
data = sim.simulate()
log.info('Results: ' + str(data))
File process.py:
# coding=utf8
"""
This file contains classes for simulating, controlling and observing a fridge.
@author: Stefan Scherfke
@contact: stefan.scherfke at uni-oldenburg.de
"""
from math import exp
import logging
import random
from SimPy.Simulation import Process, Simulation, \
activate, hold, initialize, now, simulate
log = logging.getLogger('Processes')
class Fridge(Process):
"""
This class represents a simulated fridge.
It's temperature T for and equidistant series of time steps is computed by
$T_{i+1} = \epsilon \cdot T_i + (1 - \epsilon) \cdot \left(T^O - \eta
\cdot \frac{q_i}{A}\right)$ with $\epsilon = e^{-\frac{\tau A}{m_c}}$.
"""
def __init__(self, sim, T_O = 20.0, A = 3.21, m_c = 15.97, tau = 1.0/60,
eta = 3.0, q_i = 0.0, q_max = 70.0,
T_i = 5.0, T_range = [5.0, 8.0], noise = False):
"""
Init all required variables.
@param sim: The SimPy simulation this process belongs to
@type sim: SimPy.Simulation
@param T_O: Outside temperature
@param A: Insulation
@param m_c: Thermal mass/thermal storage capacity
@param tau: Time span between t_i and t_{i+1}
@param eta: Efficiency of the cooling device
@param q_i: Initial/current electrical power
@param q_max: Power required during cool-down
@param T_i: Initial/current temperature
@param T_range: Allowed range for T_i
@param noise: Add noise to the fridge's parameters, if True
@type noise: bool
"""
Process.__init__(self, sim = sim)
self.T_O = T_O
self.A = A
self.m_c = random.normalvariate(20, 4.5) if noise else m_c
self.tau = tau
self.eta = eta
self.q_i = q_i
self.q_max = q_max
self.T_i = random.uniform(T_range[0], T_range[1]) if noise else T_i
self.T_range = T_range
def run(self):
"""
Calculate the fridge's temperature for the current time step.
"""
while True:
epsilon = exp(-(self.tau * self.A) / self.m_c)
self.T_i = epsilon * self.T_i + (1 - epsilon) \
* (self.T_O - self.eta * (self.q_i / self.A))
if self.T_i >= self.T_range[1]:
self.q_i = self.q_max # Cool down
elif self.T_i <= self.T_range[0]:
self.q_i = 0.0 # Stop cooling
log.debug('T_i: %2.2f°C at %.2f' % (self.T_i, self.sim.now()))
yield hold, self, self.tau
def coolDown(self):
"""
Start cooling down now!
"""
self.q_i = self.q_max
class FridgeObserver(Process):
"""
This process observes the temperature and power consumption of a set of
fridges.
"""
def __init__(self, sim, fridges, tau, aggSteps):
"""
Init the observer.
@param sim: The SimPy simulation this process belongs to
@type sim: SimPy.Simulation
@param fridges: A list of fridges to be observed
@type fridges: tuple of Fridge
@param tau: Time interval for observations
@type tau: float
@param aggSteps: Specifies after how many timesteps tau the collected
data is aggregated and stored.
@type aggSteps: int
"""
Process.__init__(self, sim = sim)
self._fridges = fridges
self._tau = tau
self._aggSteps = aggSteps
self._data = []
def run(self):
"""
Start observation
"""
aggSteps = 0
consumption = 0
lastProgUpdate = 0
while True:
prog = self.sim.now() * 100 / self.sim._endtime
if int(prog) > lastProgUpdate:
log.info('Progress: %d%%' % prog)
lastProgUpdate = prog
if (aggSteps >= self._aggSteps):
log.debug('Aggregating at %.2f' % self.sim.now())
self._data.append(consumption/self._aggSteps)
consumption = 0
aggSteps = 0
for fridge in self._fridges:
consumption += fridge.q_i
aggSteps += 1
yield hold, self, self._tau
def getData(self):
"""
Return the collected data
@return: a list with the collected data
"""
return self._data
if __name__ == '__main__':
logging.basicConfig(
level = logging.DEBUG,
format = '%(levelname)-8s %(asctime)s %(name)s: %(message)s')
tau = 1./60 # Step size 1min
aggSteps = 15 # Aggregate consumption in 15min blocks
params = {'tau': tau}
sim = Simulation()
fridge = Fridge(sim, **params)
observer = FridgeObserver(sim, [fridge], tau, aggSteps)
sim.activate(fridge, fridge.run(), at = 0)
sim.activate(observer, observer.run(), at = 0)
sim.simulate(until = 4 + tau)
print observer.getData()
