def fitness(self, C): """Calculate the fitness of chromosome C. The fitness of a solution is its cost. First convert C to the full representation. Then calculate its cost. (The lower the better). """ solution = copy.deepcopy(self.scenario) helpers.from_chromosome(solution, C) if self.optimize_for == "availability": return helpers.get_solution_availability(solution) elif self.optimize_for == "latency": return helpers.get_solution_global_latency(solution, self.processing_latency) else: #cost return helpers.get_solution_cost(solution)
def execute(self): """Run the genetic algorithm. Returns the scenario JSON object with placement decisions plus some algorithm-specific information. """ # Fill in missing edges helpers.add_missing_links(self.scenario) # check if we will store data per generation in a file log_generation_data = False if self.generations_file is not None: try: gen_fp = open(self.generations_file, "w+") log_generation_data = True gen_fp.write("# Scenario: " + str(self.scenario_file) + "\n") gen_fp.write("# Seed: " + str(self.seed) + "\n") gen_fp.write( "#-----------------------------------------------\n") gen_fp.write("# Generation\tFitness\t\tTimestamp\n") except: logging.warn("Error opening/writing at " + self.generations_file) pass prev_obj_value = 100000000 #inf if self.convergence_check: remaining_generations = self.stop_after start_time = datetime.now() self.init_solution_pool() for i in range(0, self.generations): obj_value = self.generation() # get a timestamp for this generation dt = datetime.now() - start_time # convert to seconds. dt.days should really not matter... time_taken = dt.days * 24 * 3600 + dt.seconds + dt.microseconds / 1000000.0 logging.info("Generation/fitness (" + self.optimize_for + ")/timestamp: " + str(i) + "\t" + str(obj_value) + "\t" + str(time_taken)) if log_generation_data: gen_fp.write( str(i) + "\t\t" + str(obj_value) + "\t" + str(time_taken) + "\n") # if we're checking for convergence to finish execution faster # we have to do some checks if self.convergence_check: if abs(obj_value - prev_obj_value) < self.delta: # the solution fitness did not significantly changed remaining_generations -= 1 else: remaining_generations = self.stop_after # the algorithm converged if remaining_generations < 0: break prev_obj_value = obj_value final_solution = helpers.from_chromosome(self.scenario, self.solution_pool[0]) # add extra information about solution performance (cost, availability, latency, time taken, # generations) # and indications about constraint violations info = self.get_solution_info(final_solution) info["generations"] = i + 1 info["execution_time"] = time_taken info["link_capacity_constraints_ok"] = True info["delay_constraints_ok"] = True info["host_capacity_constraints_ok"] = True info["mec_constraints_ok"] = True info["legal_placement"] = True # some final checks if not helpers.check_mec_constraints(final_solution): logging.warn("Final solution violates MEC constraints") info["mec_constraints_ok"] = False if not helpers.check_location_constraints(final_solution): logging.warn("Final solution violates location constraints") info["location_constraints_ok"] = False if not helpers.check_link_capacity_constraints(final_solution): logging.warn("Final solution violates link capacity constraints") info["link_capacity_constraints_ok"] = False if not helpers.check_delay_constraints(final_solution): logging.warn("Final solution violates delay constraints") info["delay_constraints_ok"] = False for h in final_solution["hosts"]: if not helpers.check_host_capacity_constraint(final_solution, h): logging.warn("Final solution violates host " + h["host_name"] + " capacity constraints") info["host_capacity_constraints_ok"] = False for v in final_solution["vnfs"]: if not helpers.check_if_placement_allowed( final_solution, v["place_at"][0], v["vnf_name"]): logging.warn("Final solution includes illegal placement of " + v["place_at"][0] + " at " + v["vnf_name"]) info["legal_placement"] = False final_solution["solution_performance"] = info used_hosts = helpers.get_used_hosts(final_solution) logging.info("Used hosts:") for uh in used_hosts: logging.info(uh) used_hedges = helpers.get_used_host_links(final_solution) logging.info("Used host edges:") for ue in used_hedges: logging.info(ue["source"] + " -> " + ue["target"] + " (" + str(ue["delay"]) + ")") # Add host edge mapping info to VNF edges helpers.add_vnf_edge_mapping(final_solution) return final_solution
def mutation(self): """Mutation operator. For each chromosome in the solution pool, decide according to the mutation rate if we'll modify it or not. If its selected for mutation, we create a mutant as follows: We select two random hosts and swap two random VNFs. If none of the selected hosts has VNFs on it, we select two other hosts and so on. If the constraints are violated, the mutant is rejected. """ counter = 0 for s in self.solution_pool: if uniform(0, 1) <= self.mutation_rate: logging.debug("Mutating solution: " + str(s)) # create a copy of the chromosome scopy = copy.deepcopy(s) # Corner-case: There's just one gene in the chromosomes, so nothing to # mutate if len(scopy.genes) < 2: continue # pick two hosts while True: h1 = choice(scopy.genes) h2 = choice(scopy.genes) if h1 == h2: continue if h1.vnfs or h2.vnfs: break # pick one VNF from each host v1 = None v2 = None if h1.vnfs: v1 = choice(h1.vnfs) h1.vnfs = [ x for x in h1.vnfs if x["vnf_name"] != v1["vnf_name"] ] if h2.vnfs: v2 = choice(h2.vnfs) h2.vnfs = [ x for x in h2.vnfs if x["vnf_name"] != v2["vnf_name"] ] # swap the two VNFs if v2: h1.vnfs.append(v2) if v1: h2.vnfs.append(v1) # create a solution represented in the full format S = helpers.from_chromosome(self.scenario, scopy) reject = False # check constraints for v in S["vnfs"]: hname = v["place_at"][0] vname = v["vnf_name"] if not helpers.check_if_placement_allowed(S, hname, vname): # There's a VNF "illegally" placed reject = True break if not reject: if helpers.check_mec_constraints(S) is False: reject = True if not reject: if helpers.check_location_constraints(S) is False: reject = True if not reject: for h in S["hosts"]: if helpers.check_host_capacity_constraint(S, h) is False: reject = True break if not reject: if helpers.check_link_capacity_constraints(S) is False: reject = True if not reject: if helpers.check_delay_constraints(S) is False: reject = True mutant = helpers.to_chromosome(S) if not reject: # all constraints ok # delete old solution and replace with mutant self.solution_pool[counter] = mutant logging.debug("Mutant ACCEPTED") else: logging.debug("Mutant REJECTED") pass counter += 1
def crossover(self): """Crossover operation. - Select two random chromosomes - Rank their genes according to an efficiency function - Create a new chromosome by taking the "best" genes until all VNFs are placed (if when adding a gene one of its VNFs is already placed, just ignore it) """ # Pick two random chromosomes (C1 and C2 could coincide) C1 = choice(self.solution_pool) C2 = choice(self.solution_pool) # Create a list of all their genes (i.e., hosts with the VNFs assigned to them) genes = copy.deepcopy(C1.genes + C2.genes) # sort genes by efficiency (lowest cost first) for g in genes: g.efficiency = self.gene_efficiency(g) # For availability, sort in descending order (as we want the max here) rev = False if self.optimize_for == "availability": rev = True genes = sorted(genes, key=attrgetter('efficiency'), reverse=rev) # vnfs to place (list of strings) vnfs = [v["vnf_name"] for v in self.scenario["vnfs"]] # create new chromosome new_genes = [] # continue as long as there are still vnfs to place while vnfs and genes: # if the gene host has already been put in the chromosome, # skip the gene. This ensures that at this step no capacity # constraints will be violated. if genes[0].hostname in [g.hostname for g in new_genes]: del (genes[0]) else: # if a VNF of the gene is not in the remaining vnf list, remove it from the gene # since this means it's already placed for v in genes[0].vnfs: if v["vnf_name"] not in vnfs: genes[0].vnfs.remove(v) # finally, add the new gene # also, remove its vnfs from the list of pending ones (there should be a more efficient way to do this) new_genes.append( Gene(genes[0].hostname, genes[0].vnfs, host_failure_rate=genes[0].host_failure_rate)) for v in genes[0].vnfs: if v["vnf_name"] in vnfs: vnfs.remove(v["vnf_name"]) del (genes[0]) C = Chromosome(new_genes) # Now we need to check if there are any VNFs left unassigned # If so, we place them anywhere they fit and are allowed to solution = copy.deepcopy(self.scenario) helpers.from_chromosome(solution, C) while vnfs: vname = vnfs.pop() v = filter(lambda x: x.get("vnf_name") == vname, solution["vnfs"])[0] host = helpers.check_if_there_is_space(solution, v) if host: # host found, place VNF v["place_at"].append(host["host_name"]) else: # Normally we should not arrive here, but, if so, # this means that there's nowhere to place the VNF # in this case, we return None and the caller will see what to do return None # perform constraint checks mec_constraints_ok = helpers.check_mec_constraints(solution) location_constraints_ok = helpers.check_location_constraints(solution) link_constraints_ok = helpers.check_link_capacity_constraints(solution) delay_constraints_ok = helpers.check_delay_constraints(solution) if link_constraints_ok and delay_constraints_ok and mec_constraints_ok and location_constraints_ok: # return the chromosome return helpers.to_chromosome(solution) else: return None