def weighted_content_placement(topology, contents, source_weights, seed=None):
"""Places content objects to source nodes randomly according to the weight
of the source node.
Parameters
----------
topology : Topology
The topology object
contents : iterable
Iterable of content objects
source_weights : dict
Dict mapping nodes nodes of the topology which are content sources and
the weight according to which content placement decision is made.
Returns
-------
cache_placement : dict
Dictionary mapping content objects to source nodes
Notes
-----
A deterministic placement of objects (e.g., for reproducing results) can be
achieved by using a fix seed value
"""
random.seed(seed)
norm_factor = float(sum(source_weights.values()))
source_pdf = dict((k, v / norm_factor) for k, v in source_weights.items())
content_placement = collections.defaultdict(set)
for c in contents:
content_placement[random_from_pdf(source_pdf)].add(c)
apply_content_placement(content_placement, topology)
def set_capacities_random(topology, capacity_pdf, capacity_unit='Mbps'):
"""
Set random link capacities according to a given probability density
function
Parameters
----------
topology : Topology
The topology to which link capacities will be set
capacity_pdf : dict
A dictionary representing the probability that a capacity value is
assigned to a link
capacity_unit : str, optional
The unit in which capacity value is expressed (e.g. Mbps, Gbps etc..)
links : list, optional
List of links, represented as (u, v) tuples to which capacity will be
set. If None or not specified, the capacity will be applied to all
links.
Examples
--------
>>> import fnss
>>> topology = fnss.erdos_renyi_topology(50, 0.1)
>>> pdf = {10: 0.5, 100: 0.2, 1000: 0.3}
>>> fnss.set_capacities_constant(topology, pdf, 'Mbps')
"""
if not capacity_unit in capacity_units:
raise ValueError("The capacity_unit argument is not valid")
if any((capacity < 0 for capacity in capacity_pdf.keys())):
raise ValueError('All capacities in capacity_pdf must be positive')
topology.graph['capacity_unit'] = capacity_unit
for u, v in topology.edges():
topology.adj[u][v]['capacity'] = random_from_pdf(capacity_pdf)
return
def weighted_content_placement(topology, contents, source_weights, seed=None):
"""Places content objects to source nodes randomly according to the weight
of the source node.
Parameters
----------
topology : Topology
The topology object
contents : iterable
Iterable of content objects
source_weights : dict
Dict mapping nodes nodes of the topology which are content sources and
the weight according to which content placement decision is made.
Returns
-------
cache_placement : dict
Dictionary mapping content objects to source nodes
Notes
-----
A deterministic placement of objects (e.g., for reproducing results) can be
achieved by using a fix seed value
"""
random.seed(seed)
norm_factor = float(sum(source_weights.values()))
source_pdf = dict((k, v / norm_factor) for k, v in source_weights.items())
content_placement = collections.defaultdict(set)
for c in contents:
content_placement[random_from_pdf(source_pdf)].add(c)
apply_content_placement(content_placement, topology)
def set_capacities_random(topology, capacity_pdf, capacity_unit='Mbps'):
"""
Set random link capacities according to a given probability density
function
Parameters
----------
topology : Topology
The topology to which link capacities will be set
capacity_pdf : dict
A dictionary representing the probability that a capacity value is
assigned to a link
capacity_unit : str, optional
The unit in which capacity value is expressed (e.g. Mbps, Gbps etc..)
links : list, optional
List of links, represented as (u, v) tuples to which capacity will be
set. If None or not specified, the capacity will be applied to all
links.
Examples
--------
>>> import fnss
>>> topology = fnss.erdos_renyi_topology(50, 0.1)
>>> pdf = {10: 0.5, 100: 0.2, 1000: 0.3}
>>> fnss.set_capacities_constant(topology, pdf, 'Mbps')
"""
if not capacity_unit in capacity_units:
raise ValueError("The capacity_unit argument is not valid")
if any((capacity < 0 for capacity in capacity_pdf.keys())):
raise ValueError('All capacities in capacity_pdf must be positive')
topology.graph['capacity_unit'] = capacity_unit
for u, v in topology.edges():
topology.adj[u][v]['capacity'] = random_from_pdf(capacity_pdf)
return
def add_m_links(G, pi):
"""Add m links between existing nodes to the graph"""
n_nodes = G.number_of_nodes()
n_edges = G.number_of_edges()
max_n_edges = (n_nodes * (n_nodes - 1)) / 2
if n_edges + m > max_n_edges: # cannot add m links
add_node(G, pi) # add a new node instead
# return in any case because before doing another operation
# (add node or links) we need to recalculate pi
return
new_links = 0
while new_links < m:
u = random_from_pdf(pi)
v = random_from_pdf(pi)
if u != v and not G.has_edge(u, v):
G.add_edge(u, v)
new_links += 1
def add_m_links(G, pi):
"""Add m links between existing nodes to the graph"""
n_nodes = G.number_of_nodes()
n_edges = G.number_of_edges()
max_n_edges = (n_nodes * (n_nodes - 1)) / 2
if n_edges + m > max_n_edges: # cannot add m links
add_node(G, pi) # add a new node instead
# return in any case because before doing another operation
# (add node or links) we need to recalculate pi
return
new_links = 0
while new_links < m:
u = random_from_pdf(pi)
v = random_from_pdf(pi)
if u != v and not G.has_edge(u, v):
G.add_edge(u, v)
new_links += 1
def add_node(G, pi):
"""Add one node to the graph and connect it to m existing nodes"""
new_node = G.number_of_nodes()
G.add_node(new_node)
new_links = 0
while new_links < m:
existing_node = random_from_pdf(pi)
if not G.has_edge(new_node, existing_node):
G.add_edge(new_node, existing_node)
new_links += 1
def add_node(G, pi):
"""Add one node to the graph and connect it to m existing nodes"""
new_node = G.number_of_nodes()
G.add_node(new_node)
new_links = 0
while new_links < m:
existing_node = random_from_pdf(pi)
if not G.has_edge(new_node, existing_node):
G.add_edge(new_node, existing_node)
new_links += 1
def extended_barabasi_albert_topology(n, m, m0, p, q, seed=None):
r"""
Return a random topology using the extended Barabasi-Albert preferential
attachment model.
Differently from the original Barabasi-Albert model, this model takes into
account the presence of local events, such as the addition of new links or
the rewiring of existing links.
More precisely, the Barabasi-Albert topology is built as follows. First, a
topology with *m0* isolated nodes is created. Then, at each step:
with probability *p* add *m* new links between existing nodes, selected
with probability:
.. math::
\Pi(i) = \frac{deg(i) + 1}{\sum_{v \in V} (deg(v) + 1)}
with probability *q* rewire *m* links. Each link to be rewired is selected as
follows: a node i is randomly selected and a link is randomly removed from
it. The node i is then connected to a new node randomly selected with
probability :math:`\Pi(i)`,
with probability :math:`1-p-q` add a new node and attach it to m nodes of
the existing topology selected with probability :math:`\Pi(i)`
Repeat the previous step until the topology comprises n nodes in total.
Parameters
----------
n : int
Number of nodes
m : int
Number of edges to attach from a new node to existing nodes
m0 : int
Number of edges initially attached to the network
p : float
The probability that new links are added
q : float
The probability that existing links are rewired
seed : int, optional
Seed for random number generator (default=None).
Returns
-------
G : Topology
References
----------
.. [1] A. L. Barabasi and R. Albert "Topology of evolving networks: local
events and universality", Physical Review Letters 85(24), 2000.
"""
def calc_pi(G):
"""Calculate extended-BA Pi function for all nodes of the graph"""
degree = G.degree()
den = float(sum(degree.values()) + G.number_of_nodes())
return {node: (degree[node] + 1) / den for node in G.nodes_iter()}
# input parameters
if n < 1 or m < 1 or m0 < 1:
raise ValueError('n, m and m0 must be a positive integer')
if m >= m0:
raise ValueError('m must be <= m0')
if n < m0:
raise ValueError('n must be > m0')
if p > 1 or p < 0:
raise ValueError('p must be included between 0 and 1')
if q > 1 or q < 0:
raise ValueError('q must be included between 0 and 1')
if p + q > 1:
raise ValueError('p + q must be <= 1')
if seed is not None:
random.seed(seed)
G = Topology(type='extended_ba')
G.name = "ext_ba_topology(%d, %d, %d, %f, %f)" % (n, m, m0, p, q)
# Step 1: Add m0 isolated nodes
G.add_nodes_from(range(m0))
while G.number_of_nodes() < n:
pi = calc_pi(G)
r = random.random()
if r <= p:
# add m new links with probability p
n_nodes = G.number_of_nodes()
n_edges = G.number_of_edges()
max_n_edges = (n_nodes * (n_nodes - 1)) / 2
if n_edges + m > max_n_edges: # cannot add m links
continue # rewire or add nodes
new_links = 0
while new_links < m:
u = random_from_pdf(pi)
v = random_from_pdf(pi)
if u is not v and not G.has_edge(u, v):
G.add_edge(u, v)
new_links += 1
elif r > p and r <= p + q:
# rewire m links with probability q
rewired_links = 0
while rewired_links < m:
i = random.choice(G.nodes()) # pick up node randomly (uniform)
if len(G.edge[i]) is 0: # if i has no edges, I cannot rewire
break
j = random.choice(list(
G.edge[i].keys())) # node to be disconnected
k = random_from_pdf(pi) # new node to be connected
if i is not k and j is not k and not G.has_edge(i, k):
G.remove_edge(i, j)
G.add_edge(i, k)
rewired_links += 1
else:
# add a new node with probability 1 - p - q
new_node = G.number_of_nodes()
G.add_node(new_node)
new_links = 0
while new_links < m:
existing_node = random_from_pdf(pi)
if not G.has_edge(new_node, existing_node):
G.add_edge(new_node, existing_node)
new_links += 1
return G
def barabasi_albert_topology(n, m, m0, seed=None):
r"""
Return a random topology using Barabasi-Albert preferential attachment
model.
A topology of n nodes is grown by attaching new nodes each with m links
that are preferentially attached to existing nodes with high degree.
More precisely, the Barabasi-Albert topology is built as follows. First, a
line topology with m0 nodes is created. Then at each step, one node is
added and connected to m existing nodes. These nodes are selected randomly
with probability
.. math::
\Pi(i) = \frac{deg(i)}{sum_{v \in V} deg V}.
Where i is the selected node and V is the set of nodes of the graph.
Parameters
----------
n : int
Number of nodes
m : int
Number of edges to attach from a new node to existing nodes
m0 : int
Number of nodes initially attached to the network
seed : int, optional
Seed for random number generator (default=None).
Returns
-------
G : Topology
Notes
-----
The initialization is a graph with with m nodes connected by :math:`m -1`
edges.
It does not use the Barabasi-Albert method provided by NetworkX because it
does not allow to specify *m0* parameter.
There are no disconnected subgraphs in the topology.
References
----------
.. [1] A. L. Barabasi and R. Albert "Emergence of scaling in
random networks", Science 286, pp 509-512, 1999.
"""
def calc_pi(G):
"""Calculate BA Pi function for all nodes of the graph"""
degree = G.degree()
den = float(sum(degree.values()))
return {node: degree[node] / den for node in G.nodes_iter()}
# input parameters
if n < 1 or m < 1 or m0 < 1:
raise ValueError('n, m and m0 must be positive integers')
if m >= m0:
raise ValueError('m must be <= m0')
if n < m0:
raise ValueError('n must be > m0')
if seed is not None:
random.seed(seed)
# Step 1: Add m0 nodes. These nodes are interconnected together
# because otherwise they will end up isolated at the end
G = Topology(nx.path_graph(m0))
G.name = "ba_topology(%d,%d,%d)" % (n, m, m0)
G.graph['type'] = 'ba'
# Step 2: Add one node and connect it with m links
while G.number_of_nodes() < n:
pi = calc_pi(G)
u = G.number_of_nodes()
G.add_node(u)
new_links = 0
while new_links < m:
v = random_from_pdf(pi)
if not G.has_edge(u, v):
G.add_edge(u, v)
new_links += 1
return G
def extended_barabasi_albert_topology(n, m, m0, p, q, seed=None):
r"""
Return a random topology using the extended Barabasi-Albert preferential
attachment model.
Differently from the original Barabasi-Albert model, this model takes into
account the presence of local events, such as the addition of new links or
the rewiring of existing links.
More precisely, the Barabasi-Albert topology is built as follows. First, a
topology with *m0* isolated nodes is created. Then, at each step:
with probability *p* add *m* new links between existing nodes, selected
with probability:
.. math::
\Pi(i) = \frac{deg(i) + 1}{\sum_{v \in V} (deg(v) + 1)}
with probability *q* rewire *m* links. Each link to be rewired is selected as
follows: a node i is randomly selected and a link is randomly removed from
it. The node i is then connected to a new node randomly selected with
probability :math:`\Pi(i)`,
with probability :math:`1-p-q` add a new node and attach it to m nodes of
the existing topology selected with probability :math:`\Pi(i)`
Repeat the previous step until the topology comprises n nodes in total.
Parameters
----------
n : int
Number of nodes
m : int
Number of edges to attach from a new node to existing nodes
m0 : int
Number of edges initially attached to the network
p : float
The probability that new links are added
q : float
The probability that existing links are rewired
seed : int, optional
Seed for random number generator (default=None).
Returns
-------
G : Topology
References
----------
.. [1] A. L. Barabasi and R. Albert "Topology of evolving networks: local
events and universality", Physical Review Letters 85(24), 2000.
"""
def calc_pi(G):
"""Calculate extended-BA Pi function for all nodes of the graph"""
degree = dict(G.degree())
den = float(sum(degree.values()) + G.number_of_nodes())
return {node: (degree[node] + 1) / den for node in G.nodes()}
# input parameters
if n < 1 or m < 1 or m0 < 1:
raise ValueError('n, m and m0 must be a positive integer')
if m >= m0:
raise ValueError('m must be <= m0')
if n < m0:
raise ValueError('n must be > m0')
if p > 1 or p < 0:
raise ValueError('p must be included between 0 and 1')
if q > 1 or q < 0:
raise ValueError('q must be included between 0 and 1')
if p + q > 1:
raise ValueError('p + q must be <= 1')
if seed is not None:
random.seed(seed)
G = Topology(type='extended_ba')
G.name = "ext_ba_topology(%d, %d, %d, %f, %f)" % (n, m, m0, p, q)
# Step 1: Add m0 isolated nodes
G.add_nodes_from(range(m0))
while G.number_of_nodes() < n:
pi = calc_pi(G)
r = random.random()
if r <= p:
# add m new links with probability p
n_nodes = G.number_of_nodes()
n_edges = G.number_of_edges()
max_n_edges = (n_nodes * (n_nodes - 1)) / 2
if n_edges + m > max_n_edges: # cannot add m links
continue # rewire or add nodes
new_links = 0
while new_links < m:
u = random_from_pdf(pi)
v = random_from_pdf(pi)
if u is not v and not G.has_edge(u, v):
G.add_edge(u, v)
new_links += 1
elif r > p and r <= p + q:
# rewire m links with probability q
rewired_links = 0
while rewired_links < m:
i = random.choice(list(G.nodes())) # pick up node randomly (uniform)
if len(G.adj[i]) is 0: # if i has no edges, I cannot rewire
break
j = random.choice(list(G.adj[i].keys())) # node to be disconnected
k = random_from_pdf(pi) # new node to be connected
if i is not k and j is not k and not G.has_edge(i, k):
G.remove_edge(i, j)
G.add_edge(i, k)
rewired_links += 1
else:
# add a new node with probability 1 - p - q
new_node = G.number_of_nodes()
G.add_node(new_node)
new_links = 0
while new_links < m:
existing_node = random_from_pdf(pi)
if not G.has_edge(new_node, existing_node):
G.add_edge(new_node, existing_node)
new_links += 1
return G
def barabasi_albert_topology(n, m, m0, seed=None):
r"""
Return a random topology using Barabasi-Albert preferential attachment
model.
A topology of n nodes is grown by attaching new nodes each with m links
that are preferentially attached to existing nodes with high degree.
More precisely, the Barabasi-Albert topology is built as follows. First, a
line topology with m0 nodes is created. Then at each step, one node is
added and connected to m existing nodes. These nodes are selected randomly
with probability
.. math::
\Pi(i) = \frac{deg(i)}{sum_{v \in V} deg V}.
Where i is the selected node and V is the set of nodes of the graph.
Parameters
----------
n : int
Number of nodes
m : int
Number of edges to attach from a new node to existing nodes
m0 : int
Number of nodes initially attached to the network
seed : int, optional
Seed for random number generator (default=None).
Returns
-------
G : Topology
Notes
-----
The initialization is a graph with with m nodes connected by :math:`m -1`
edges.
It does not use the Barabasi-Albert method provided by NetworkX because it
does not allow to specify *m0* parameter.
There are no disconnected subgraphs in the topology.
References
----------
.. [1] A. L. Barabasi and R. Albert "Emergence of scaling in
random networks", Science 286, pp 509-512, 1999.
"""
def calc_pi(G):
"""Calculate BA Pi function for all nodes of the graph"""
degree = dict(G.degree())
den = float(sum(degree.values()))
return {node: degree[node] / den for node in G.nodes()}
# input parameters
if n < 1 or m < 1 or m0 < 1:
raise ValueError('n, m and m0 must be positive integers')
if m >= m0:
raise ValueError('m must be <= m0')
if n < m0:
raise ValueError('n must be > m0')
if seed is not None:
random.seed(seed)
# Step 1: Add m0 nodes. These nodes are interconnected together
# because otherwise they will end up isolated at the end
G = Topology(nx.path_graph(m0))
G.name = "ba_topology(%d,%d,%d)" % (n, m, m0)
G.graph['type'] = 'ba'
# Step 2: Add one node and connect it with m links
while G.number_of_nodes() < n:
pi = calc_pi(G)
u = G.number_of_nodes()
G.add_node(u)
new_links = 0
while new_links < m:
v = random_from_pdf(pi)
if not G.has_edge(u, v):
G.add_edge(u, v)
new_links += 1
return G
def weighted_repo_content_placement(topology,
contents,
freshness_per,
shelf_life,
msg_size,
topics_weights,
types_weights,
max_replications,
source_weights,
service_weights,
max_label_nos,
seed=None):
"""Places content objects to source nodes randomly according to the weight
of the source node.
TODO: This should be modified, or another one created, to include content
placement parameters, like the freshness periods, shelf-lives, topics/types
of labels and placement possibilities, maybe depending on hashes, placement
of nodes and possibly other scenario-specific/service-specific parameters.
ADD SERVICE TYPE TO MESSAGE PROPERTIES!
Parameters
----------
topology : Topology
The topology object
contents : iterable
Iterable of content objects
topics :
types :
freshness_per :
shelf_life :
msg_size :
source_weights : dict
Dict mapping nodes of the topology which are content sources and
the weight according to which content placement decision is made.
Returns
-------
cache_placement : dict
Dictionary mapping content objects to source nodes
Notes
-----
A deterministic placement of objects (e.g., for reproducing results) can be
achieved by using a fix seed value
"""
# TODO: This is the format that each datum (message) shuold have
# placed_data = {content, msg_topics, msg_type, freshness_per,
# shelf_life, msg_size}
placed_data = dict()
random.seed(seed)
norm_factor = float(sum(source_weights.values()))
# TODO: These ^\/^\/^ might need redefining, to make label-specific
# source weights, and then the labels distributed according to these.
# OR the other way around, distributing sources according to label weights
if types_weights is not None:
types_labels_norm_factor = float(sum(types_weights.values()))
types_labels_pdf = dict((k, v / types_labels_norm_factor)
for k, v in types_weights.items())
topics_labels_norm_factor = float(sum(topics_weights.values()))
service_labels_norm_factor = float(sum(service_weights.values()))
# TODO: Think about a way to randomise, but still maintain a certain
# distribution among the users that receive data with certain labels.
# Maybe associate the pdf with labels, rather than contents, SOMEHOW!
source_pdf = dict((k, v / norm_factor) for k, v in source_weights.items())
topics_labels_pdf = dict(
(k, v / topics_labels_norm_factor) for k, v in topics_weights.items())
service_labels_pdf = dict((k, v / service_labels_norm_factor)
for k, v in service_weights.items())
service_association = collections.defaultdict(set)
labels_association = collections.defaultdict(set)
content_placement = collections.defaultdict(set)
# Further TODO: Add all the other data characteristics and maybe place
# content depending on those at a later point (create other
# placement strategies)
# NOTE: All label names will come as a list of strings
for c in contents:
alter = False
if freshness_per is not None:
if placed_data.has_key(contents[c]['content']):
placed_data[contents[c]['content']].update(
freshness_per=freshness_per)
else:
placed_data[contents[c]['content']] = dict()
placed_data[contents[c]
['content']]['freshness_per'] = freshness_per
if shelf_life is not None:
placed_data[contents[c]['content']].update(shelf_life=shelf_life)
if max_replications:
placed_data[contents[c]['content']].update(
max_replications=max_replications)
placed_data[contents[c]['content']].update(replications=0)
service_association[random_from_pdf(service_labels_pdf)].add(c)
placed_data[contents[c]['content']].update(content=c)
placed_data[contents[c]['content']].update(msg_size=msg_size)
placed_data[contents[c]['content']]["receiveTime"] = 0
placed_data[contents[c]['content']]['labels'] = contents[c]['labels']
placed_data[contents[c]['content']]['service_type'] = "non-proc"
if not placed_data[contents[c]['content']]['labels']:
for i in range(0, max_label_nos):
if types_weights is not None and not alter:
labels_association[random_from_pdf(types_labels_pdf)].add(
c)
alter = True
elif topics_weights is not None and alter:
labels_association[random_from_pdf(topics_labels_pdf)].add(
c)
alter = False
elif topics_weights is not None:
labels_association[random_from_pdf(topics_labels_pdf)].add(
c)
placed_data = apply_labels_association(labels_association, placed_data)
#placed_data = apply_service_association(service_association, placed_data)
for d in placed_data:
rand = random_from_pdf(source_pdf)
if not content_placement[rand]:
content_placement[rand] = dict()
if content_placement[rand].has_key(d):
content_placement[rand][d].update(placed_data[d])
else:
content_placement[rand][d] = dict()
content_placement[rand][d] = placed_data[d]
apply_content_placement(content_placement, topology)
topology.placed_data = placed_data