def test_sampling(self):
# FIXME: bogotest, extend tests!
dist = PDist.lognorm(0.25)
sampler = BoundRejectionSampler(dist, 0.5, 2)
x = sampler.sample(10000)
self.assertGreater(x.min(), 0.5)
self.assertLess(x.max(), 2)
def inject_workload_generator(self, env, deployment: Tuple[int, Pod, Pod, Pod]):
i, pod0, pod1, pod2 = deployment
logger.debug('%.2f injecting request generators for deployment %d', env.now, i)
prep_function_name = f'wf_0_preprocess_{i}'
training_trigger = request_generator(
env,
ParameterizedDistribution.expon(((200, 100,), None, None)),
lambda: FunctionRequest(prep_function_name, empty)
)
inference_function_name = f'wf_2_inference_{i}'
inference_trigger = request_generator(
env,
ParameterizedDistribution.expon(((25, 50,), None, None)),
lambda: FunctionRequest(inference_function_name, empty)
)
env.process(training_trigger)
env.process(inference_trigger)
import random
from srds import ParameterizedDistribution as PDist
# python-pi was made with req=10000, except xeongpu - this one was req=2000, due to the inexplicable slow response
execution_time_distributions = {
('rpi3', 'faas-workloads/python-pi'):
(1.909987211227417, 2.203219413757324,
PDist.lognorm(
((1.5561351906586467, ), 1.9092154433285176, 0.03832589013392289))),
('rpi4', 'faas-workloads/python-pi'):
(1.6961045265197754, 1.7755048274993896,
PDist.lognorm(
((2.0378973843923234, ), 1.6960468751471511, 0.017276257000211737))),
('xeongpu', 'faas-workloads/python-pi'):
(2.9558749198913574, 3.0182230472564697,
PDist.lognorm(
((0.33924794244178996, ), 2.9421626090307855, 0.03231526504686262))),
('xeoncpu', 'faas-workloads/python-pi'):
(2.9558749198913574, 3.0182230472564697,
PDist.lognorm(
((0.33924794244178996, ), 2.9421626090307855, 0.03231526504686262))),
('nx', 'faas-workloads/python-pi'):
(1.280782699584961, 1.3314659595489502,
PDist.lognorm(
((0.3204243885534094, ), 1.2658584067446368, 0.029294830565182426))),
('tx2', 'faas-workloads/python-pi'):
(0.9472050666809082, 0.973700761795044,
PDist.lognorm(
((0.34123939468038056, ), 0.9407173896320605, 0.015806120791955418))),
('nano', 'faas-workloads/python-pi'):
from srds import ParameterizedDistribution
from ether.blocks import nodes
from ether.blocks.cells import Cloudlet, IoTComputeBox, MobileConnection, FiberToExchange
from ether.cell import GeoCell, SharedLinkCell
from ether.topology import Topology
default_num_cells = 3
default_cloudlet_size = (5, 2)
default_cell_density = ParameterizedDistribution.lognorm((0.82, 2.02))
class UrbanSensingScenario:
def __init__(self,
num_cells=default_num_cells,
cell_density=default_cell_density,
cloudlet_size=default_cloudlet_size,
internet='internet') -> None:
"""
The UrbanSensingScenario builds on ideas from the Array of Things project, but extends it with proximate compute
resources and adds a cloudlet to the city.
The city is divided into cells, e.g., neighborhoods, and each cell has multiple urban sensing nodes and
proximate compute resources. The devices in a cell are connected via a shared link. The city also hosts a
cloudlet composed of server computers.
The high-level parameters are: the number of cells, the cell density (number of nodes per cell), and the
cloudlet size.
:param num_cells: the number of cells to create, e.g., the neighborhoods in a city
:param cell_density: the distribution describing the number of nodes in each neighborhood
from srds import ParameterizedDistribution as PDist
execution_time_distributions = {
('cloud', 'alexrashed/ml-wf-1-pre'):
(0.584, 1.1420000000000001,
PDist.lognorm(
((0.31449780857108944, ), 0.36909628354997315, 0.41583220283981315))),
('tegra', 'alexrashed/ml-wf-1-pre'):
(0.434, 0.491,
PDist.lognorm(
((0.4040891723912467, ), 0.42388853817361616, 0.026861281861394234))),
('pi', 'alexrashed/ml-wf-1-pre'):
(23.89, 108.743,
PDist.lognorm(
((0.20506023311489027, ), -20.283277542855338, 73.25412435048207))),
('nuc', 'alexrashed/ml-wf-1-pre'): (0.10099999999999999, 0.105,
PDist.lognorm(
((23.561803013427763, ),
0.10099999999999998,
0.00012383535141814733))),
('cloud', 'alexrashed/ml-wf-2-train'):
(156.256, 166.702,
PDist.lognorm(
((0.0955189827829152, ), 140.38073254297746, 20.5799433435967))),
('tegra', 'alexrashed/ml-wf-2-train'):
(17.641, 20.333,
PDist.lognorm(
((0.07553515034079655, ), 12.501696816822736, 6.339535271582099))),
('nuc', 'alexrashed/ml-wf-2-train'):
(31.326999999999998, 42.355,
PDist.lognorm(
from srds import ParameterizedDistribution as PDist lan = PDist.lognorm((0.25, 0.35, 0.16)) wlan = PDist.lognorm((0.635, 1.18, 3.27)) business_isp = PDist.lognorm((0.87, 5.95, 1.21)) mobile_isp = PDist.lognorm((0.49, 16.2, 8.02))
def exp_sampler(lambd):
sampler = BufferedSampler(PDist.expon(lambd))
while True:
yield sampler.sample()