def action_map(self, _range=1):
self.actions = set()
for i in range(_range + 1):
a = i * 100 / settings.ACTIONS
Mlog.INFO("ACTION: ", a)
self.actions.add(a)
self.actions.add(-a)
self.actions = tuple(self.actions)
Mlog.INFO('THE ACTIONS: ', self.actions)
return self.actions
def get_reward(self, clients, old_action):
""" recuperar a recompensa para uma mudança de estado """
reward = 0
gain1 = self.client.get_gain1()
#gain2 = self.client.get_gain2()
self.client.update()
#print(self.dc.load, self.dc.cap, self.actions[self.old_action])
if settings.STRATEGY in ('fql_step', 'fsl_step'):
old_action = self.old_action
else:
old_action = self.actions[self.old_action]
if self.dc.load >= self.dc.cap:
Mlog.INFO("SERVER GREATER", self.dc.load, self.dc.cap)
if old_action >= 0:
reward = -3 #*(self.dc.load/self.dc.cap)
return reward
else:
reward = 1 #* (self.dc.load / self.dc.cap)
return reward
if self.client.nbw < 0.5 * self.client.bw:
#Mlog.INFO('OLD ACTION', self.client.id, old_action)
if old_action >= 0:
reward = 0.7 #*self.client.nbw/self.client.bw
else:
#Mlog.INFO("NEGATIVE REWARD BY CLIENT UNDER SLA")
reward = -4 #*self.client.nbw/self.client.bw
return reward
if self.client.nbw < self.client.bw:
if old_action >= 0:
#print()
reward = 0.5 #* self.client.nbw / self.client.bw
return reward
else:
#Mlog.INFO("NEGATIVE REWARD BY CLIENT UNDER SLA 2")
reward = -2 #* self.client.nbw / self.client.bw
return reward
#if self.SLA_PROBLEM:
#return -0.7*self.client.nbw/self.client.bw
return -0.7 * self.client.nbw / self.client.bw
reward += gain1
if gain1 < 1:
return -reward
return reward
def get_current_state(self):
if settings.STATE_TYPE == 'val':
# OLD
state = int(self.dc.load / (self.dc.cap / self.OBSERVATION_SPACE))
elif settings.STATE_TYPE == 'diff':
state = int(
self.dc.load /
(self.old_dc_load * self.dc.cap / self.OBSERVATION_SPACE))
self.old_dc_load = self.dc.load
elif settings.STATE_TYPE == 'fuzzy':
cpu = psutil.cpu_percent()
load = self.dc.load
state = int(
self.fls.get_state(cpu=psutil.cpu_percent(), bw=self.dc.load))
Mlog.INFO('FUZZY INPUT: ', cpu, load, state)
if state > self.OBSERVATION_SPACE:
return self.dump_state(self.OBSERVATION_SPACE)
return self.dump_state(state)