A novel approach of gateway
selection and placement in cellular
1. Establishing metro-scale “cellular Wi-Fi” network to support
seamless Internet access in the urban area.
2. Find the minimum number of gateways and their optimal
placement for a given network graph so as to minimize the
network installation costs while maintaining reliability,
flexibility, and an acceptable grade of service.
3. Develop a set of linear inequalities based on various constraints.
4. Solve the Integer Linear Programming(ILP) model by using
IEEE 802.11 has one of the fastest adaptation rate seen in the
history of technology.
In the near future Access Points (APs) will be scattered over an
entire city enabling people to use any mobile devices equipped with
IEEE 802.11 network interface card.
The cellular Wi-Fi system is essentially a meshed infrastructure
network, which is different from the existing cellular systems and
ad hoc networks.
Only a minimum number of APs called gateways in cellular Wi-Fi
are connected to the wired backbone, while other APs connect to
the gateways through single or multi-hop wireless links.
Related Work and our Contribution
In most of the clustering techniques used in multihop wireless
network, the cluster structure is controlled by hop distance.
Graph Partitioning: The graph partitioning problem is NP-
Complete. Partitioning software’s make best effort to partition
the graph in k-parts, where each part has equal number of nodes.
Most algorithm to select the location of gateways to interconnect
a mobile data network with a fixed data network based on
minimizing average packet delays.
Different gateways have different bandwidth capacity, and each
gateway can serve different number of APs.
Our algorithm gives Good cluster
Any AP with a connection to the backbone network (e.g. via DSL,
cable modem, T1, T3 or OC1 line) is a candidate for gateway.
Constraints of an efficient gateway
1. Topology independent: The algorithm should be independent of
network structure, with the ability of working efficiently for any
given network topology.
2. Transparency: The algorithm should not impact the client and
should be totally transparent to users.
3. Load balancing: It should distribute the network load as evenly
as possible among the gateways.
4. Fault tolerance: It should prevent single point of failure.
Parameters of ILP model
Proposed ILP model
Explanation of ILP model (1)
node ‘j’ is associated with node ‘i’, if node ‘i’ has been
selected as the gateway and node ‘j’ is an AP
Explanation of ILP model (2)
3. To make sure that the total traffic of all the APs selected by a
gateway is not more than the maximum capacity of the gateway.
4. Once a node has been selected, it cannot be selected by
Explanation of ILP model (3)
Cost Function :We define the cost by considering both the
establishment cost of the gateways and number of hops its
associated APs are away from the gateway.
Minimize the overall system cost:
Result of our algorithm
Result of greedy approach
Population is uniformly distributed with a population density of
3371 per square mile by using the city of Houston, TX example.
25% of people use Wi-Fi devices and subscribe to cellular Wi-Fi
Approximately 10 APs are needed per square mile so each AP will
serve approximately 84 people.
A normal user can withstand a delay of 10 seconds for his web
page to be loaded and he roughly takes 30 seconds to absorb the
content of the page.
A normal user will not require more than 20 Kbps for normal web
surfing, email, citrix , messaging systems and the like.
Gateway capacity varies from 1.544 Mbps to 51.85 Mbps.
The inequalities discussed in ILP model are solved by using
lpsolve, a simplex-based code for linear and integer programming
problems by Michel Berkelaar.
The results obtained from lpsolve are compared with greedy
algorithm for gateway selection.
The results are discussed in the following slides
We assumed that 10% of the nodes in the network graph are the
gateway candidates. Every mobile device sharing an AP can
simultaneously connect at 20 Kbps. Number of nodes varies
from 25 to 100.
The total number of node is 60 and each user generates data
traffic at a rate of 25kbps.The percentage of candidate nodes
varies from 5% to 20%.
The bandwidth usage by each user is varied from 20kbps to
45kbps. The total number of nodes is fixed at 60 and 10%
of nodes are potential candidate for gateway.
The proposed approach can effectively identify a minimum
number of gateways at optimal locations in a cellular Wi-Fi
network, resulting in significant lower cost and better
The proposed ILP model considers not only the capacity of
the gateways, but also economic aspects associated with
installing them and minimum hop count of all the associated
APs with a particular gateway.
The algorithm provides a high degree of reliability, flexibility
It also prevents single point of failure.