PROFITABLE CONNECTION
ASSIGNMENT IN ALL OPTICAL
WDM NETWORKS
VISHAL ANAND
LANDER
(Lab. for Advanced Network Design, Evaluation and Research)
In collaboration with:
Tushar Katarki and Chunming Qiao
CSE Dept., SUNY at Buffalo
Outline
Introduction
Related work
Maximum Profitability Problem
Concluding remarks
Questions and discussion
2
Vishal Anand
Introduction
Optical WDM networks - future backbone
for wide area networks.
Physical Topology - Optical wavelength
routers connected by fiber links.
Lightpath or connection - Path between
two end nodes and a wavelength on that
path.
No wavelength conversion - Any lightpath
uses the same wavelength on all the links
its path spans.
Vishal Anand 3
The RWA(Routing & Wavelength
Assignment) Problem
Given :
– a network topology
– a set of traffic demands (or connection
requests).
Determine the routes and wavelengths to use
so as to satisfy the demands.
The RWA problem is usually solved to
optimize some specified objective(s).
Vishal Anand 4
Previous and related work
Example objectives
– Minimize network-wide packet delay (e.g.
number of hops).
– Maximize throughput (e.g. number of
lightpaths).
– Maximize allowable capacity upgrade or
scalability (for future traffic demands).
Minimizing cost (network resources used)
can also be an important objective.
For a bandwidth broker (or carrier)
maximizing profits is most important.
Vishal Anand 5
The Maximum Profitability Problem
Given:
– a set of connection requests, N.
– a network topology.
– earnings (revenue) Ei associated with each
connection request, i.
– cost of using any wavelength on a link l, Cl.
Solve the RWA problem to maximize the
profit, P = Total Earnings - Total Costs.
Vishal Anand 6
Maximizing profit problem is a more
general formulation.
– If Ei=E, for each connection/lightpath i (i.e., all
connections have equal earnings) OR if n=N
(i.e., all the connection requests have to be
satisfied) then the problem is same as the
minimizing cost problem.
– If Ei=E and if all connections/lightpaths have
equal costs. Then the problem is the same as
maximizing throughput problem.
Hence a more direct study of the
maximizing profit problem is necessary.
Vishal Anand 7
Network Model
Network topology considered: 16 node
NSFNET.
Cost of using each wavelength on a link , is
the same, but varies from link to link.
No wavelength conversion capabilities at
any of the nodes.
Number of wavelengths on each link in the
network is the same.
8
Vishal Anand
Heuristic based approach
The RWA problem is known to be NP Hard
and hence computationally intractable.
Maximizing profit heuristic: MaxPro
– Find a cheapest path for each connection request
and compute the profit.
– Sort the requests in the order of decreasing
profit and store in a list.
– Satisfy connection requests in decreasing order
of profit (a greedy approach).
– If a connection request is satisfied.
• delete that connection from the sorted list.
9
Vishal Anand
– If the cheapest path for a connection request is
not available.
• Re-compute a new cheapest path for only that
connection request.
• Compute the new profit for this connection request.
• Insert this connection into the sorted list depending
on the profit.
– Repeat till no other connection request can be
satisfied.
10
Vishal Anand
Results and Comparison
1) Results obtained from MaxPro compared
with:
– a minimizing cost heuristic
– a random assignment heuristic
2) Results of Maxpro compared with the
optimal results from integer linear program.
11
Vishal Anand
Comparison of the heuristics
MaxPro performs the best.
Better than a minimizing cost heuristic.
12
Vishal Anand
Integer Linear Model
Definitions:
– W :Number of wavelengths on each link.
– n, L :Number of nodes, links in the network.
– :Earnings obtained by satisfying a connection request between
nodes and .
– : Total number of alternate routes/paths available to reach node
from node .
– : Cost of reaching node from node on route r.
– : = 1, if link j is used by the route r between nodes and , 0
otherwise.
– : Number of connection requests between nodes and .
– : = 1, indicates that the connection between node and is
routed on route r using wavelength k.
13
Vishal Anand
Objective function
Subject to:
– The total number of lightpaths established between a node pair
should not exceed the number of requests between that node pair.
14
Vishal Anand
And
– The total number of lightpaths established on any link should not
exceed the number of wavelength on that link.
– A wavelength on a link can support at most one lightpath.
– The Integrality constraint.
15
Vishal Anand
Comparison of Maxpro with ILP
MaxPro obtains on the average 90% of the
results got from the ILP(optimal profit).
16
Vishal Anand
Summary and Future work
Formulated a RWA problem with the
objective of maximizing the profit.
Proposed a maximizing profit heuristic.
Compared results of a profit maximizing
heuristic with a minimizing cost heuristic
and ILP.
Future work
– Study the maximizing profit problem for the
On-line traffic model.
– Extend to cases where protection and
restoration is required for the traffic.
17
Vishal Anand