Source code for fnss.topologies.datacenter

"""
Provides methods to generate commonly adopted datacenter topologies.

Each topology generation function returns an instance of DatacenterTopology
"""

import networkx as nx
from fnss.topologies.topology import Topology


__all__ = ['DatacenterTopology',
           'two_tier_topology',
           'three_tier_topology',
           'bcube_topology',
           'fat_tree_topology'
           ]


[docs]class DatacenterTopology(Topology): """ Represent a datacenter topology """
[docs] def number_of_switches(self): """ Return the number of switches in the topology """ return len(self.switches())
[docs] def number_of_servers(self): """ Return the number of servers in the topology """ return len(self.servers())
[docs] def switches(self): """ Return the list of switch nodes in the topology """ return [v for v in self.nodes() if self.node[v]['type'] == 'switch']
[docs] def servers(self): """ Return the list of server nodes in the topology """ return [v for v in self.nodes() if self.node[v]['type'] == 'server']
[docs]def two_tier_topology(n_core, n_edge, n_servers): """ Return a two-tier datacenter topology. This topology comprises switches organized in two tiers (core and edge) and servers connected to edge routers. Each core switch is connected to each edge switch while each server is connected to exactly one edge switch. Each node has two attributes: * type: can either be *switch* or *server* * tier: can either be *core*, *edge* or *leaf*. Nodes in the leaf tier are only server, while all core and edge nodes are switches. Each edge has an attribute type as well which can either be 'core_edge' if it connects a core and an edge switch or 'edge_leaf' if it connects an edge switch to a server. Parameters ---------- n_core : int Total number of core switches n_edge : int Total number of edge switches n_servers : int Number of servers connected to each edge switch. Returns ------- topology : DatacenterTopology """ # validate input arguments if type(n_core) is not int or type(n_edge) is not int \ or type(n_servers) is not int: raise TypeError('n_core, n_edge and n_servers must be integers') if n_core < 1 or n_edge < 1 or n_servers < 1: raise ValueError('n_core, n_edge and n_servers must be positive') topo = DatacenterTopology(nx.complete_bipartite_graph(n_core, n_edge)) topo.name = "two_tier_topology(%d,%d,%d)" % (n_core, n_edge, n_servers) topo.graph['type'] = 'two_tier' for u in range(n_core): topo.node[u]['tier'] = 'core' topo.node[u]['type'] = 'switch' for v in topo.edge[u]: topo.edge[u][v]['type'] = 'core_edge' for u in range(n_core, n_core + n_edge): topo.node[u]['tier'] = 'edge' topo.node[u]['type'] = 'switch' for _ in range(n_servers): v = topo.number_of_nodes() topo.add_node(v) topo.node[v]['tier'] = 'leaf' topo.node[v]['type'] = 'server' topo.add_edge(u, v, type='edge_leaf') return topo
[docs]def three_tier_topology(n_core, n_aggregation, n_edge, n_servers): """ Return a three-tier data center topology. This topology comprises switches organized in three tiers (core, aggregation and edge) and servers connected to edge routers. Each core switch is connected to each aggregation, each edge switch is connected to one aggregation switch and finally each server is connected to exactly one edge switch. Each node has two attributes: * type: can either be *switch* or *server* * tier: can either be *core*, *aggregation*, *edge* or *leaf*. Nodes in the leaf tier are only server, while all core, aggregation and edge nodes are switches. Each edge has an attribute type as well which can either be *core_edge* if it connects a core and an aggregation switch, *aggregation_edge*, if it connects an aggregation and a core switch or *edge_leaf* if it connects an edge switch to a server. The total number of servers is :math:`n_aggregation * n_edge * n_servers`. Parameters ---------- n_core : int Total number of core switches n_aggregation : int Total number of aggregation switches n_edge : int Number of edge switches per each each aggregation switch n_servers : int Number of servers connected to each edge switch. Returns ------- topology : DatacenterTopology """ # Validate input arguments if type(n_core) is not int or type(n_aggregation) is not int \ or type(n_edge) is not int \ or type(n_servers) is not int: raise TypeError('n_core, n_edge, n_aggregation and n_servers '\ 'must be integers') if n_core < 1 or n_aggregation < 1 or n_edge < 1 or n_servers < 1: raise ValueError('n_core, n_aggregation, n_edge and n_server '\ 'must be positive') topo = DatacenterTopology(nx.complete_bipartite_graph(n_core, n_aggregation)) topo.name = "three_tier_topology(%d,%d,%d,%d)" % (n_core, n_aggregation, n_edge, n_servers) topo.graph['type'] = 'three_tier' for u in range(n_core): topo.node[u]['tier'] = 'core' topo.node[u]['type'] = 'switch' for v in topo.edge[u]: topo.edge[u][v]['type'] = 'core_aggregation' for u in range(n_core, n_core + n_aggregation): topo.node[u]['tier'] = 'aggregation' topo.node[u]['type'] = 'switch' for _ in range(n_edge): v = topo.number_of_nodes() topo.add_node(v) topo.node[v]['tier'] = 'edge' topo.node[v]['type'] = 'switch' topo.add_edge(u, v, type='aggregation_edge') total_n_edge = topo.number_of_nodes() for u in range(n_core + n_aggregation, total_n_edge): for _ in range(n_servers): v = topo.number_of_nodes() topo.add_node(v) topo.node[v]['tier'] = 'leaf' topo.node[v]['type'] = 'server' topo.add_edge(u, v, type='edge_leaf') return topo
[docs]def bcube_topology(n, k): """ Return a Bcube datacenter topology, as described in [1]_: The BCube topology is a topology specifically designed for shipping-container based, modular data centers. A BCube topology comprises hosts with multiple network interfaces connected to commodity switches. It has the peculiar characteristic that switches are never directly connected to each other and servers are used also for packet forwarding. This topology is defined as a recursive structure. A :math:`Bcube_0` is composed of n servers connected to an n-port switch. A :math:`Bcube_1` is composed of n :math:`Bcube_0` connected to n n-port switches. A Bcube_k is composed of n :math:`Bcube_{k-1}` connected to :math:`n^k` n-port switches. This topology comprises: * :math:`n^(k+1)` servers, each of them connected to :math:`k+1` switches * :math:`n*(k+1)` switches, each of them having n ports Each node has an attribute type which can either be 'switch' or 'server' and an attribute 'level' which specifies at what level of the Bcube hierarchy it is located Each edge also has the attribute *level*. Parameters ---------- k : int The level of Bcube n : int The number of server per Bcube_0 Returns ------- topology : DatacenterTopology References ---------- .. [1] C. Guo, G. Lu, D. Li, H. Wu, X. Zhang, Y. Shi, C. Tian, Y. Zhang, and S. Lu. BCube: a high performance, server-centric network architecture for modular data centers. Proceedings of the ACM SIGCOMM 2009 conference on Data communication (SIGCOMM '09). ACM, New York, NY, USA. http://doi.acm.org/10.1145/1592568.1592577 """ # Validate input arguments if type(n) is not int or type(k) is not int: raise TypeError('k and h arguments must be of int type') if n < 1: raise ValueError("Invalid n parameter. It should be >= 1") if k < 0: raise ValueError("Invalid k parameter. It should be >= 0") topo = DatacenterTopology(type='bcube') topo.name = "bcube_topology(%d,%d)" % (n, k) # add servers n_servers = n**(k + 1) topo.add_nodes_from(range(n_servers), type='server') # add all layers of switches and connect them to servers for level in range(k + 1): # i is the horizontal position of a switch a specific level for i in range(n**k): u = topo.number_of_nodes() # add switch at given level topo.add_node(u, level=level, type='switch') servers = range(i, i + n**(level + 1), n**level) for v in servers: topo.add_edge(u, v, level=level) return topo
[docs]def fat_tree_topology(k): """ Return a fat tree datacenter topology, as described in [1]_ A fat tree topology built using k-port switches can support up to :math:`(k^3)/4` hosts. This topology comprises k pods with two layers of :math:`k/2` switches each. In each pod, each aggregation switch is connected to all the :math:`k/2` edge switches and each edge switch is connected to :math:`k/2` hosts. There are :math:`(k/2)^2` core switches, each of them connected to one aggregation switch per pod. Each node has three attributes: * type: can either be *switch* or *server* * tier: can either be *core*, *aggregation*, *edge* or *leaf*. Nodes in * pod: the pod id in which the node is located, unless it is a core switch the leaf tier are only server, while all core, aggregation and edge nodes are switches. Each edge has an attribute type as well which can either be *core_edge* if it connects a core and an aggregation switch, *aggregation_edge*, if it connects an aggregation and a core switch or *edge_leaf* if it connects an edge switch to a server. Parameters ---------- k : int The number of ports of the switches Returns ------- topology : DatacenterTopology References ---------- .. [1] M. Al-Fares, A. Loukissas, and A. Vahdat. A scalable, commodity data center network architecture. Proceedings of the ACM SIGCOMM 2008 conference on Data communication (SIGCOMM '08). ACM, New York, NY, USA http://doi.acm.org/10.1145/1402958.1402967 """ # validate input arguments if type(k) is not int: raise TypeError('k argument must be of int type') if k < 1 or k % 2 == 1: raise ValueError('k must be a positive even integer') topo = DatacenterTopology(type='fat_tree') topo.name = "fat_tree_topology(%d)" % (k) # Create core nodes n_core = (k//2)**2 topo.add_nodes_from([v for v in range(int(n_core))], layer='core', type='switch') # Create aggregation and edge nodes and connect them for pod in range(k): aggr_start_node = topo.number_of_nodes() aggr_end_node = aggr_start_node + k//2 edge_start_node = aggr_end_node edge_end_node = edge_start_node + k//2 aggr_nodes = range(aggr_start_node, aggr_end_node) edge_nodes = range(edge_start_node, edge_end_node) topo.add_nodes_from(aggr_nodes, layer='aggregation', type='switch', pod=pod) topo.add_nodes_from(edge_nodes, layer='edge', type='switch', pod=pod) topo.add_edges_from([(u, v) for u in aggr_nodes for v in edge_nodes], type='aggregation_edge') # Connect core switches to aggregation switches for core_node in range(n_core): for pod in range(k): aggr_node = n_core + (core_node//(k//2)) + (k*pod) topo.add_edge(core_node, aggr_node, type='core_aggregation') # Create servers and connect them to edge switches for u in [node for node in topo.nodes() if topo.node[node]['layer'] == 'edge']: leaf_nodes = range(topo.number_of_nodes(), topo.number_of_nodes() + k//2) topo.add_nodes_from(leaf_nodes, layer='leaf', type='server', pod=topo.node[u]['pod']) topo.add_edges_from([(u, v) for v in leaf_nodes], type='edge_leaf') return topo