def calculate_initial_state(instances, T):
"""Calculates the initial values for the given instances about total cost, total capacity and instance types
:param instances: instances.
:type instances: list.
:param T: running time in hours for each instance, there is one by one mapping with first argument.
:type statistics: list.
:returns: total cost, total capacity and instance types.
"""
capacity = cloudData.get_instances_capacity(instances)
prices = cloudData.get_instances_cost(instances)
cost = 0.0
for i in range(len(instances)):
cost = cost + T[i]*prices[i]
types=[]
for instance in instances:
types.append({'os': instance['os'] , 'region': instance['region'] , 'type': instance['type'] })
total_capacity = sum(capacity)
return cost, total_capacity, types
def cost_utilization_optimization(instances, statistics, operation='Average', f=1, W=0.5):
"""Algorithm for optimizing the cost and the utilization of a set of instances
:param instances: instances.
:type instances: list.
:param statistics: static and dynamic data regarding the given instances.
:type statistics: list.
:param f: parameter determining the targeted utilization (f=1 for 100%).
:type f: double.
:param W: importance given to the cost (W) in relation to the utilization (1-W), 0<=W<=1
:type W: double.
:returns: something.
:raises: ExecutionError
"""
if len(instances) != len(statistics):
raise Exceptions.ExecutionError("cost_utilization_optimization","Instances and statistics lists have different size")
os_vectors = utils.create_os_vectors()
wl = cloudData.get_instances_workload(instances,statistics,operation)
m = len(wl)
(D,capacity,cost,T) = get_instances_info(instances,statistics)
cost_vector = []
for i in range(len(T)):
for cj in range(len(cost)):
cost_vector.append(cost[cj]*T[i])
initial_cost = 0
instances_cost = cloudData.get_instances_cost(instances)
for i in range(len(instances)):
initial_cost = initial_cost+instances_cost[i]*T[i]
DU = sum(wl)/f
my_obj = [0]*m*D
my_obj.append(1-W)
my_obj.append(W)
my_colnames = []
for i in range(m):
for j in range(D):
my_colnames.append("x"+str(i)+"_"+str(j))
my_colnames.extend("U"+"C")
my_rhs = []
my_rhs1 = [1]*m
my_rhs2 = []
for i in range(m):
my_rhs2.extend(capacity)
my_rhs3 = [0,0,DU]
my_rhs = my_rhs1+my_rhs2+my_rhs3
my_type = []
for i in range(m*D):
my_type.extend("I")
my_type.extend("C"*2)
num_c = m+m*D+2+1
my_rownames = []
for i in range(num_c):
my_rownames.append("c"+str(i+1))
my_sense = []
my_sense.extend("E"*m)
my_sense.extend("L"*(m*D))
my_sense.extend("E"*2)
my_sense.extend("G")
def populatebyrow(prob):
prob.objective.set_sense(prob.objective.sense.minimize)
prob.variables.add(obj = my_obj, names = my_colnames, types = my_type)
rows = []
rows1 = []
rows2 = []
my_rows = []
for i in range(m):
rows1 = my_colnames[i*D:i*D+D]
if instances[i]['os'] == 'linux':
rows2 = os_vectors[1]
else:
rows2 = os_vectors[0]
rows.append(rows1)
rows.append(rows2)
my_rows.append(rows)
rows = []
list2=[]
for i in range(m):
for j in range(D):
list2 = [[my_colnames[i*D+j]],[wl[i]]]
my_rows.append(list2)
list2 = []
list3 = []
list4 = []
list3 = [my_colnames[0:len(my_colnames)-1], capacity*m+[-1.0]]
list4 = [my_colnames[0:len(my_colnames)-2] + [my_colnames[len(my_colnames)-1]], cost_vector+[-1.0] ]
my_rows.append(list3)
my_rows.append(list4)
list5 = [["U"],[ 1.0]]
my_rows.append(list5)
prob.linear_constraints.add(lin_expr = my_rows, senses = my_sense,rhs = my_rhs, names = my_rownames)
my_prob = cplex.Cplex()
my_prob.parameters.read.datacheck.set(1)
handle = populatebyrow(my_prob)
my_prob.set_results_stream(None)
my_prob.solve()
numrows = my_prob.linear_constraints.get_num()
numcols = my_prob.variables.get_num()
cplex_solution = { "code": my_prob.solution.get_status() , "status": str(my_prob.solution.status[my_prob.solution.get_status()]) , "value": my_prob.solution.get_objective_value() }
numcols = my_prob.variables.get_num()
numrows = my_prob.linear_constraints.get_num()
slack = my_prob.solution.get_linear_slacks()
x = my_prob.solution.get_values()
cuo_instance_types = []
all_types = cloudData.get_all_instance_types()
for j in range(numcols):
if x[j]==1:
cuo_instance_types.append(all_types[j%D])
(initial_cost,initial_capacity,initial_instance_types) = calculate_initial_state(instances,T)
return { "initial_cost": initial_cost , "cuo_cost": x[numcols-1] , "initial_total_capacity":initial_capacity , "cuo_total_capacity":x[numcols-2] , "total_workload": sum(wl) , "initial_instance_types": initial_instance_types , "cuo_instance_types": cuo_instance_types }
