def make_containers(self): #workload
#assumption is that a service is equivalent to a set of containers
#all tasks in a service have common requirements because they are replicas
#service specifications
tasks_per_service = [
np.random.randint(1, 4) for i in range(self.total_services)
] #add randomness
for i in range(self.total_services):
self.service_cpu.append(
np.random.randint(self.service_cpu_low, self.service_cpu_high))
self.service_mem.append(
np.random.randint(self.service_mem_low, self.service_mem_high))
#make containers
containers_kub = [] #Instantiated as objects of kubernetes.Container
containers_ga = [] #Instantiated as objects of ga.Container
id = 0
for i in range(len(tasks_per_service)):
for j in range(tasks_per_service[i]):
containers_kub.append(
kubernetes.Container(self.service_cpu[i],
self.service_mem[i], str(id)))
containers_ga.append(
ga.Container(self.service_cpu[i], self.service_mem[i],
str(id)))
id += 1
return (containers_kub, containers_ga)
def off_1_test():
n1 = ga.Node (4, 16000, 4, 16000, 140, 80)
n2 = ga.Node (4, 8000, 4, 8000, 160, 90)
n3 = ga.Node (2, 1024, 2, 1024, 180, 100)
n4 = ga.Node (6, 6512, 6, 6512, 200, 110)
nodes_info = [n1,n2,n3,n4]
node_ids = [0, 0, 0, 0, 2, 2, 3, None]
c1 = ga.Container(1, 128, "A")
c2 = ga.Container(1, 128, "B")
c3 = ga.Container(1, 128, "C")
c4 = ga.Container(1, 128, "D")
c5 = ga.Container(1, 128, "E")
c6 = ga.Container(1, 128, "F")
c7 = ga.Container(1, 128, "G")
c8 = ga.Container(1, 128, "H")
containers = [c1, c2, c3, c4, c5, c6, c7, c8]
ex1 = ga.Chromosome(node_ids, containers, nodes_info)
n1.containers_list = [c1, c2, c3, c4]
n3.containers_list = [c5, c6]
n4.containers_list = [c7]
return ex1.get_fitness()
def off_3_test():
n1 = ga.Node (4, 16000, 4, 16000, 140, 80)
n2 = ga.Node (4, 8000, 4, 8000, 160, 90)
n3 = ga.Node (6, 10024, 2, 1024, 180, 100)
n4 = ga.Node (6, 6512, 6, 6512, 200, 110)
nodes_info = [n1,n2,n3,n4]
node_ids = [0, 0, 0, 3, 2, 2, 2, 2, 1, 3, 1, 1, 3]
c1 = ga.Container(1, 128, "A")
c2 = ga.Container(1, 128, "A")
c3 = ga.Container(1, 128, "A")
c4 = ga.Container(1, 128, "A")
c5 = ga.Container(1, 128, "D")
c6 = ga.Container(1, 128, "D")
c7 = ga.Container(1, 128, "D")
c8 = ga.Container(1, 128, "D")
c9 = ga.Container(1, 128, "C")
c10 = ga.Container(1, 128, "C")
c11 = ga.Container(1, 128, "B")
c12 = ga.Container(1, 128, "B")
c13 = ga.Container(1, 128, "B")
containers = [c1, c2, c3, c4, c5, c6, c7, c8, c9, c10, c11, c12, c13]
ex1 = ga.Chromosome(node_ids, containers, nodes_info)
n1.containers_list = [c1, c2, c3]
n2.containers_list = [c9, c11, c12]
n3.containers_list = [c5, c6, c7, c8]
n4.containers_list = [c4, c10, c13]
return ex1.get_fitness()
def main():
#1)make physical configs
#2)decide workload
#3)get allocation from kub
#4)get allocation from nsga
#5)measure quality of kub allocation vs nsga allocation
#6)show graphs of kub vs final nsga
#7)look at history of nsga and make graphs of front
start = timeit.timeit()
#AWS A1
per_type_A1 = 20
cpu_A1 = [1, 2, 4, 8, 16]
mem_A1 = [2048, 4096, 8192, 16384, 32768]
#AWS M4
per_type_M4 = 40
cpu_M4 = [2, 4, 8, 16, 40, 64]
mem_M4 = [8192, 16384, 32768, 65536, 163840, 262144]
#AWS M5a
per_type_M5a = 80
cpu_M5a = [2, 4, 8, 16, 32, 48, 64, 96]
mem_M5a = [8192, 16384, 32768, 65536, 131072, 196608, 262144, 393216]
#Parameters for instantiating an Experiment are node_cpu,node_mem,per_type,service_cpu,service_mem,total_services,service_cpu_low,service_cpu_high,service_mem_low,service_mem_high
Exp_1 = Experiment(cpu_A1, mem_A1, per_type_A1, [], [], 100, 0, 2, 25,
1000)
Exp_2 = Experiment(cpu_M4, mem_M4, per_type_M4, [], [], 200, 1, 2, 50,
2000)
Exp_3 = Experiment(cpu_M5a, mem_M5a, per_type_M5a, [], [], 400, 2, 4, 200,
4000)
Exp_4 = Experiment(cpu_M5a, mem_M5a, per_type_M5a, [], [], 100, 0, 1, 25,
1000)
Experiments = [Exp_1] #[Exp_1, Exp_2, Exp_3, Exp_4]
for experiment in Experiments:
(nodes_kub,
nodes_ga) = experiment.make_nodes() #list of nodes of Node class
(containers_kub, containers_ga) = experiment.make_containers(
) #list of containers of Container class
kub_alloc = get_kub_allocations(nodes_kub, containers_kub)
chr = ga.Chromosome(find_assigned(kub_alloc), containers_kub,
kub_alloc, False, None)
(fa, fb, fc, fd,
fe) = chr.get_fitness() #fitness of kubernetes allocation
nsga_alloc, history = get_nsga_allocations(nodes_ga, containers_ga)
visualize_history(history, "5_obj.png")
(f1, f2, f3, f4,
f5) = nsga_alloc.get_fitness() #fitness of selected nsga solution
x_names = ['Obj1', 'Obj2', 'Obj3', 'Obj4', 'Obj5']
kub = [fa, fb, fc, fd, fe]
nsga = [f1, f2, f3, f4, f5]
visualise.obj_over_configs(
x_names, kub, nsga, "Objective Values",
"Perfomance of Kubernetes vs NSGA-3 on the 5 fitness objectives")
end = timeit.timeit()
print("Execution Time: %s" % (end - start))
#Rescheduling Experiment
#Adding dummy container, so that rescheduling is needed
nodes = nsga_alloc.nodes_info
min_cpu = 0
min_mem = 0
for node in nodes:
if node.remaining_cpu < experiment.service_cpu_high and node.remaining_memory < experiment.service_mem_high:
min_cpu = experiment.service_cpu_high
min_mem = experiment.service_mem_high
dummy_container = ga.Container(min_cpu, min_mem,
len(containers_ga))
containers_ga.append(dummy_container)
#nsga_alloc.node_ids.append(node.id)
#node.containers_list.append(dummy_container) #maybe then need to compute obj6 except last element
break
nsga_rescheduling, history = get_nsga_rescheduling(
nodes_ga, containers_ga, nsga_alloc)
visualize_history(history, '6_obj.png')
(off1, off2, off3, off4, off5, off6) = nsga_rescheduling.get_fitness()
#Rescheduling impemented as Scheduling
nsga_re_scheduling, _ = get_nsga_allocations(nodes_ga, containers_ga)
off_6 = nsga_re_scheduling.off_6(nsga_alloc)
(off_1, off_2, off_3, off_4, off_5) = nsga_re_scheduling.get_fitness()
x_names = ['Obj1', 'Obj2', 'Obj3', 'Obj4', 'Obj5', 'Obj6']
resch = [off1, off2, off3, off4, off5, off6]
re_sch = [off_1, off_2, off_3, off_4, off_5, off_6]
visualise.obj_over_configs_rescheduling(
x_names, resch, re_sch, "Objective Values",
"Perfomance of Rescheduling on the 6 fitness objectives")