def new_cartesius_pilot():
pilot = Pilot()
pilot.resource_name = "cartesius"
pilot.node_count = 1
pilot.time = 1 | units.hour
pilot.queue_name = "short"
pilot.slots_per_node = 24
pilot.label = "hydro"
return pilot
def __init__(self):
print("Setting up distributed code")
instance = DistributedAmuse()
instance.initialize_code()
# instance.parameters.debug = True
# instance.parameters.webinterface_port = 4556
instance.commit_parameters()
print("url:", instance.get_webinterface_url())
print("Resources:")
print(instance.resources)
#Claim nodes on the resources. In this example simply the "local" machine
pilot = Pilot()
pilot.resource_name = 'local' # label of the resource to be used
pilot.node_count = 1 # desired number of nodes
pilot.time = 99 | units.hour # wallclock that resource remains available (mainly for job queues)
pilot.slots_per_node = 99 # slots is accounting measure for a job
pilot.node_label = 'local' # label for subgroups of the resource
instance.pilots.add_pilot(pilot)
print("Reservations:")
print(instance.pilots)
print("Waiting for reservations")
instance.wait_for_pilots()
self.instance = instance
def init_local_only():
print("Setting up distributed code")
instance = DistributedAmuse()
instance.initialize_code()
instance.parameters.webinterface_port = 4556
print("url:", instance.get_webinterface_url())
#~ print "Resources:"
#~ print instance.resources
#Claim nodes on the resources. In this example simply the "local" machine
pilot = Pilot()
pilot.resource_name = 'local'
pilot.node_count = 1
pilot.time = 99 | units.hour
pilot.slots_per_node = 99
pilot.node_label = 'local'
instance.pilots.add_pilot(pilot)
#~ print "Reservations:"
#~ print instance.pilots
print("Waiting for reservations")
instance.wait_for_pilots()
# atexit.register(instance.stop)
return instance
def setup_class(cls):
cls.check_not_in_mpiexec()
super(TestCDistributedImplementationInterface, cls).setup_class()
#~ print "Setting up distributed code"
#instance = DistributedAmuse(redirection='none')
cls.distinstance = cls.new_instance_of_an_optional_code(
DistributedAmuse) #, redirection='none')
cls.distinstance.parameters.debug = False
#~ print "Resources:"
#~ print cls.distinstance.resources
pilot = Pilot()
pilot.resource_name = 'local'
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = 2
pilot.label = 'local'
cls.distinstance.pilots.add_pilot(pilot)
#~ print "Pilots:"
#~ print cls.distinstance.pilots
#~ print "Waiting for pilots"
cls.distinstance.wait_for_pilots()
cls.distinstance.use_for_all_workers()
def setUp(self):
self.check_not_in_mpiexec()
super(TestCDistributedImplementationInterface, self).setUp()
print "Setting up distributed code"
#instance = DistributedAmuse(redirection='none')
self.distinstance = self.new_instance_of_an_optional_code(
DistributedAmuse) #, redirection='none')
self.distinstance.parameters.debug = False
print "Resources:"
print self.distinstance.resources
pilot = Pilot()
pilot.resource_name = 'local'
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = 2
pilot.label = 'local'
self.distinstance.pilots.add_pilot(pilot)
print "Pilots:"
print self.distinstance.pilots
print "Waiting for pilots"
self.distinstance.wait_for_pilots()
self.distinstance.use_for_all_workers()
def new_hofvijver_pilot():
pilot = Pilot()
pilot.resource_name = "hofvijver"
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = 64
pilot.label = "hydro"
return pilot
def new_cpu_node_pilot(resource):
pilot = Pilot()
pilot.resource_name = resource.name
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = 10
pilot.label = "CPU"
return pilot
def new_gpu_node_pilot(resource, slots=2, label="GPU"):
pilot = Pilot()
pilot.resource_name = resource.name
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = slots
pilot.label = label
return pilot
def new_local_pilot(slots=10, label="local"):
pilot = Pilot()
pilot.resource_name = "local"
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = slots
pilot.label = label
return pilot
def new_local_pilot():
pilot = Pilot()
pilot.resource_name = "local"
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = 10
pilot.node_label = "local"
return pilot
def init_das5_only(username, num_nodes, num_cores):
print("Setting up distributed code")
instance = DistributedAmuse()
instance.parameters.debug = False
instance.parameters.worker_queue_timeout=1 | units.hour
instance.parameters.webinterface_port = 4556
print("url:", instance.get_webinterface_url())
instance.commit_parameters()
#print "Resources:"
resource = Resource()
resource.name = "DAS-5"
resource.location = username + "@fs0.das5.cs.vu.nl"
resource.scheduler_type = "slurm"
resource.amuse_dir = "/home/" + username + "/amuse/amuse"
resource.tmp_dir = "/home/" + username + "/tmp"
instance.resources.add_resource(resource)
#print instance.resources
pilot = Pilot()
pilot.resource_name="DAS-5"
pilot.queue_name="defq"
pilot.node_count=num_nodes
pilot.time= 24|units.hour
pilot.slots_per_node=num_cores
pilot.label="DAS-5-Pilot"
instance.pilots.add_pilot(pilot)
#~ print "Reservations:"
#~ print instance.pilots
print("Waiting for reservations")
instance.wait_for_pilots()
return instance
def start_distributed_amuse():
print "Creating distributed amuse"
distributed_amuse = DistributedAmuse(redirection='none')
distributed_amuse.parameters.debug = True
distributed_amuse.parameters.webinterface_port = 4556
distributed_amuse.use_for_all_workers()
#open the address of the webinterface in a brower window
webbrowser.open(distributed_amuse.get_webinterface_url())
#Add some resources
#resource = Resource()
#resource.name='some.machine'
#resource.location="*****@*****.**"
#resource.scheduler_type="sge"
#resource.amuse_dir="/home/user/amuse"
#distributed_amuse.resources.add_resource(resource)
print "Resources:"
print distributed_amuse.resources
#Claim nodes on the resources. In this example simply the "local" machine
pilot = Pilot()
pilot.resource_name = 'local'
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = 22
pilot.label = 'local'
distributed_amuse.pilots.add_pilot(pilot)
print "Pilots:"
print distributed_amuse.pilots
print "Waiting for pilots"
distributed_amuse.wait_for_pilots()
print "setting distributed as default channel"
distributed_amuse.use_for_all_workers()
return distributed_amuse
def start_distributed_amuse():
print "Creating distributed amuse"
distributed_amuse = DistributedAmuse(redirection='none')
distributed_amuse.parameters.debug = True
distributed_amuse.parameters.webinterface_port = 4556
distributed_amuse.use_for_all_workers()
#open the address of the webinterface in a brower window
webbrowser.open(distributed_amuse.get_webinterface_url())
#Add some resources
#resource = Resource()
#resource.name='some.machine'
#resource.location="*****@*****.**"
#resource.scheduler_type="sge"
#resource.amuse_dir="/home/user/amuse"
#distributed_amuse.resources.add_resource(resource)
print "Resources:"
print distributed_amuse.resources
#Claim nodes on the resources. In this example simply the "local" machine
pilot = Pilot()
pilot.resource_name='local'
pilot.node_count=1
pilot.time= 2|units.hour
pilot.slots_per_node=22
pilot.label='local'
distributed_amuse.pilots.add_pilot(pilot)
print "Pilots:"
print distributed_amuse.pilots
print "Waiting for pilots"
distributed_amuse.wait_for_pilots()
print "setting distributed as default channel"
distributed_amuse.use_for_all_workers()
return distributed_amuse
def setUp(self):
self.check_not_in_mpiexec()
super(TestCDistributedImplementationInterface, self).setUp()
print "Setting up distributed code"
#instance = DistributedAmuse(redirection='none')
self.distinstance = self.new_instance_of_an_optional_code(DistributedAmuse)#, redirection='none')
self.distinstance.parameters.debug = False
print "Resources:"
print self.distinstance.resources
pilot = Pilot()
pilot.resource_name='local'
pilot.node_count=1
pilot.time= 2|units.hour
pilot.slots_per_node=2
pilot.label='local'
self.distinstance.pilots.add_pilot(pilot)
print "Pilots:"
print self.distinstance.pilots
print "Waiting for pilots"
self.distinstance.wait_for_pilots()
self.distinstance.use_for_all_workers()
# open the address of the webinterface in a brower window # webbrowser.open(instance.get_webinterface_url()) # Add some resources # resource = Resource() # resource.name='DAS4-VU' # resource.location="*****@*****.**" # resource.scheduler_type="sge" # resource.amuse_dir="/home/user/amuse" # instance.resources.add_resource(resource) print "Resources:" print instance.resources # Claim nodes on the resources. In this example simply the "local" machine pilot = Pilot() pilot.resource_name = "local" pilot.node_count = 1 pilot.time = 2 | units.hour pilot.slots_per_node = 22 pilot.label = "local" instance.pilots.add_pilot(pilot) print "Pilots:" print instance.pilots print "Waiting for pilots" instance.wait_for_pilots() print "setting distributed as default channel" instance.use_for_all_workers()
#initialize code, print output of code to console
instance = DistributedAmuse(redirection='none')
instance.commit_parameters()
print instance.parameters.webinterface_port
resource = Resource()
resource.name = "Cartesius"
resource.location = "*****@*****.**"
resource.scheduler_type = "slurm"
resource.amuse_dir = "/home/ben/amuse/amuse-svn"
resource.tmp_dir = "/home/ben"
instance.resources.add_resource(resource)
pilot = Pilot()
pilot.resource_name="Cartesius"
pilot.queue_name="short"
pilot.node_count=1
pilot.time= 1|units.hour
pilot.slots_per_node=24
pilot.label="CartesiusNode"
instance.pilots.add_pilot(pilot)
instance.use_for_all_workers()
from omuse.community.pop.interface import POP
p=POP(channel_type="distributed", redirection="none", number_of_workers=24)
p.change_directory('/home/ben/amuse/amuse-svn/src/omuse/community/pop/')
# instance.parameters.debug = True
# instance.parameters.webinterface_port = 4556
instance.commit_parameters()
# Add some resources
# resource = Resource()
# resource.name='DAS4-VU'
# resource.location="*****@*****.**"
# resource.scheduler_type="sge"
# resource.amuse_dir="/home/user/amuse"
# instance.resources.add_resource(resource)
print("Resources:")
print(instance.resources)
# Claim nodes on the resources. In this example simply the "local" machine
pilot = Pilot()
pilot.resource_name = 'local'
pilot.node_count = 1
pilot.time = 2 | units.hour
pilot.slots_per_node = 32
pilot.label = 'local'
instance.pilots.add_pilot(pilot)
print("Pilots:")
print(instance.pilots)
print("Waiting for pilots")
instance.wait_for_pilots()
print("setting distributed as default channel")
instance.use_for_all_workers()
instance.parameters.debug = False instance.parameters.worker_queue_timeout=1 | units.hour instance.commit_parameters() print instance.parameters.webinterface_port resource = Resource() resource.name = "DAS-5" resource.location = "*****@*****.**" resource.scheduler_type = "slurm" resource.amuse_dir = "/home/bwn200/amuse/amuse-svn" resource.tmp_dir = "/home/bwn200/tmp" instance.resources.add_resource(resource) pilot = Pilot() pilot.resource_name="DAS-5" pilot.queue_name="defq" pilot.node_count=56 pilot.time= 24|units.hour pilot.slots_per_node=16 pilot.label="DAS-5-Pilot" instance.pilots.add_pilot(pilot) instance.use_for_all_workers() from omuse.community.pop.interface import POP p=POP(channel_type="distributed", redirection="none", mode='3600x2400x42', number_of_workers=896, max_message_length=1000000) #set grid info
