PERFORMANCE OF TOPOLOGY AGGREGATION SCHEMES IN

HIERARCHICAL ROUTING

A.Sherly Alphonse, E.Baburaj,

II M.E.Computer Science,
Arulmigu Kalasalingam College of Engineering,
Virudhunagar Dt. Krishnancoil,
Tamil Nadu State, India.
Phone: 04563-289042
e-mail id: sherls_82@myway.com
Lecturer, Computer Science Department,
Arulmigu Kalasalingam College of Eng.
Virudhunagar Dt. Krishnancoil,
Tamil Nadu State, India
e-mail id: alanchy_babu@yahoo.com

Abstract: As the network grows larger and larger, it becomes impossible to broadcast the whole topology to
every node in the network, as it takes up an enormous amount of space, time and bandwidth. One solution to
deal with this scalability problem is to group nodes into manageable domains and then aggregate the internal
topologies of these domains before broadcasting them to other domains. This process is known as Topology
Aggregation. Topology Aggregation often introduces distortion, that is, deviation from the original topology. It
becomes very difficult to do Topology Aggregation within bounded distortion. With this kind of distorted and
summarized view of domains available to a node, Multi-Constrained QoS routing becomes extremely difficult.
The major contribution of this paper is a novel QOS parameter representation with a new aggregation algorithm
and a QOS–aware routing protocol (Line Segment Routing Algorithm). We examine the impact of re-aggregation
policy, the criteria that trigger the re-aggregation and re-advertisement of domain topology. The QOS
representation captures the state information about the network with much greater accuracy than the existing
algorithms. The new approach achieves very good performance in terms of delay deviation, success ratio and
crank back ratio.

Keyword: hierarchical routing, QoS, topology aggregation, delay- bandwidth sensitive, lsra.

1. INTRODUCTION compaction and the impact of such a compaction

Routing consists of two main tasks. The first
task is to collect network state information and keep
it up-to-date. The second is to find a feasible path for
a new connection based on the collected information
[1]. As the network size grows, it becomes highly
unrealistic to broadcast the whole topology to every
node in the network as it will take up too much
space, time as well as bandwidth.[2] There are a few
proposed solutions to deal with this scalability
problem. Among them are: (1) Reducing the
frequency of topology updates (2) Reducing the size
of topology updates and (3) Combining these two
techniques[2]. This paper examines the topology
updates size reduction method .
on the routing performance[3][4].
When a network state is aggregated,
imprecision is introduced. A logical node may
represent a large subnet with complex internal
structure and a logical link may be an abstraction of
multiple physical links. This has a negative impact
on QoS routing. The problem becomes more
complicated when multiple QoS parameters are
involved[5]. Some paths may have better bandwidth
availability and some others may have smaller
delay[1]. There may not be a path with the best
properties in both terms. To aggregate such
information is a big problem and therefore this leads
to distortion.

Topology aggregation is achieved by All aggregation schemes suffer from

grouping neighboring nodes into routing domains
and representing the routing information in each
domain (PG) in a compact manner.fig.1, 2. An
efficient topology aggregation scheme is one that
provides a proper balance between topology
different degrees of distortion [6]. That is the cost
obtained from the aggregated topology deviates
from the original value. Hence routing has to be
based on inaccurate information and it becomes NP-
hard to do routing. There are several ways to do
aggregation with bounded distortion when there is
only one metric under consideration[7]. When more
than 1 parameters is under consideration it’s
difficult to do aggregation under bounded distortion.

2. MATERIALS AND METHODOLOGIES

First a new topology called quasi-star, is defined

which is a star-like topology whose nucleus is a
border node of a PG and whose leaves are the
remaining M - 1 border nodes. The logical link
connecting nucleus and border nodes are extracted
from mesh. These M representations are advertised
to outside through various border nodes.
a star, the PG computes the weights of links from
each border node to the fictitious nucleus, and vice
versa.
In our case, the weight of a link from a
border node ni to the fictitious nucleus f is is taken as
the average weight of logical links that originate
from this border node in the full-mesh. More
formally, from each node ni

(1)

(a) whole network (b)view of a node in domain A

fig. 1.Network Model

 (a)Domain F (b) Mesh of the borders
fig.2. Topology Aggregation
Here in this TA scheme for each
PG , the full-mesh of M border nodes is
reduced into a star whose leaves are the
border nodes. The center of the star is a fictitious
nucleus (fig.1, 2). In contrast to the default
(symmetric) representation, the links between the
nucleus and the leaves have different weights, which
can be determined in various ways [5]. In the
following scheme we will consider the average case
approach.
To aggregate a PG, the PG first constructs a
full-mesh of border nodes. The weight of each
logical link in the full-mesh is the distance between
two border nodes. To aggregate the full-mesh into
The weight of a link from the fictitious nucleus f to a
border node nj is the average weight of logical links
that end at this border node. More formally, for each
node nj, the PG computes
(2)
The complexity of the conventional
asymmetric star approach is O (M), which is a
compromise between a full-mesh and a simple node.
Nevertheless, the asymmetric star approach is
still lossy [8 ]. This approach
relies on the
observation presented above; the most
important information is provided in the
first received quasi- star[9].
The source-oriented star scheme
first finds the star representation of the
full-mesh as described above. Then, it
(c) Star representation (d) Star representation with bypasses
combines each quasi-star with the same
star and constructs M distinct complex-
node representations of the same PG.
Each representation consists of the
same star but a different quasi-star. When
an outside node receives a complex-node
representation of a PG for the first time, it
stores it and forwards it to its neighbors
except the one from which the
advertisement received.
We perform re-aggregation whenever there
is any change in the bandwidth requirement. We call
this the logarithmic update re-aggregation policy.
Total bandwidth on each link is divided into log [B]
+1 Blocks. Re-aggregation is performed only when
the bandwidth crosses these division boundaries.
2.1 How it works?
Figure 3 shows a PG in which each
physical link is associated with two QoS
parameters bandwidth (BW) and delay.
There are five distinct paths from Node A
to Node b. One path is a→1→3→b. The
end-to-end QoS parameter of this path is
(4,5).We can find the other parameters
(7,9)(10,5),(2,3) and (7,7).

Fig.3 A PG in which a link has multiple paths
We would like to represent these
five paths by a single logical link with
appropriate bandwidth and delay values
(x,y).Note that parameter (2,3) is better
than parameter (7,7) in terms of delay,
but not in terms of bandwidth. Given a set
(S) of points on the delay-bandwidth
plane, (x, y) Є S is a representative of S, if
there does not exist any point (x’, y’) Є S
which is more representative than (x, y).
V (x’,y’)Є S, x<x’ or y>y’. The dotted line
is a stair case from left to right. The
representatives are points on the convex
corners of the steps. Thus it’s enough to
store only the representative points. A
line segment is used to approximate the
staircase.
2.1.1. Line Segment Routing Algorithm
This approach brings forth a solution for
networks similar to the one addressed by Bauer in
[11]. The network with one additive and one
restrictive parameter is termed here as Bandwidth-
Delay Sensitive Networks. The (BW,delay)
parameter space is shown in Figure 4.
The basic idea is to use a line segment to
represent the efficient frontier on the Cartesian plane
for a logical link . The line segment is found using
linear regression. A sample line segment is given in
Figure 4.
Whatever request falls below the line
segment is rejected and above is admitted as
routable. Obviously, those requests that are below
the line segment but outside the dotted staircase
function will be cranked back as they are admitted
although they in the inadmissible region. Similarly,
all requests that fall in the region above the line
segment but below the staircase function are rejected
although these requests are routable through the link.
Fig. 4 The geometric representation of Line Segment
The mesh represented by line segments is
then transformed into an asymmetric star with
bypasses topology. TA method Routing algorithms
that are tailored for the line segments are developed
in the paper. Simulations are performed on static
networks and matters of re-aggregation are discussed
in the paper.
LSRA is a QoS based source routing
algorithm, integrating modified Dijkstra’s Algorithm
(DA) and the centralized bandwidth-delay routing
algorithm (CBDRA) [10]. DA is an optimal
algorithm when finding the shortest-delay path when
delay is the only metric. CBDRA works when there
are 2 metrics, delay and bandwidth. It first prunes all
the links that do not satisfy the bandwidth
requirement, and then applies DA to find the min-
delay path in the residue network.[11]. In LSRA this
idea is augmented to deal with line segment
representation.[12]. LSRA requires that each node
keeps the topology of its own PG and star
aggregations of other PGs[13][14].There are 2 levels
are routing: inter-domain and intra-domain.
LSRA accordingly has 2 phases:
A. Inter-Domain Routing
 Each node see all nodes in its own domain
and all border nodes of the other domains. There are
five steps in inter-domain routing.
1. Transform star with bypasses to meshes:
Since the nuclei of stars are virtual,the actual
routing path should not include any nucleus.
2. Prune physical links:
This step prunes the logical links
that do not satisfy the bandwidth requirement.
3. Determine the delays of logical links :
The delay value,supported by a line segment
is a function of the bandwidth requirement.This step
determines the delay values of all logical links under
the bandwidth requirement.
4. Prune physical links:
This step prunes the physical links that do
not satisfy the bandwidth requirement
5. Apply DA on the network:
This step uses DA to find the shortest delay
path to the destination domain.
B. Intra-Domain Routing
In this phase LSRA finds the route in a
distributed fashion. A message or a packet is sent
from the source to travel along the inter-domain path
found. When the border node t of PG g receives the
message, and finds out the next hop in the inter-
domain path is another border node t’ in g, it finds the
path going from t to t’ using CBDRA [15].t has the
complete knowledge about its own domain g. Node t
inserts the intradomain path into the interdomain path
that is carried by the message .Then the message is
sent to t’ along the intradomain path. Complexity of
CBDRA is same as DA.
3. COMPARISON WITH EXISTING SYSTEMS
The state of a path in a delay-
bandwidth sensitive network can be represented as a
delay-bandwidth pair to a point on the delay-
bandwidth plane. Since each link after aggregation
may be the abstraction of many physical paths, a
single delay-bandwidth pair which is a point on the
delay-bandwidth plane is not sufficient to capture
the QoS parameters of all those paths. According to
reference [8] it used a curve to approximate the
properties of all the paths between 2 border nodes
without any precedence among the parameters.
This curve had the following properties like
minimum delay, maximum bandwidth and the
smallest stretch factor among all the paths between
the border nodes. Though this provided better
approximation than a single point this had several
short comings like the routing algorithm with
polynomial complexity is not provided, instead the
algorithm is provided only to find a path whether a
path is likely to be feasible or not. In certain cases
only one QoS metric will contribute its information
and the information about the other metric will be
lost.
We examine the impact of re-aggregation
policy, the criteria that trigger the re-aggregation and
re-advertisement of domain topology. The QOS
representation captures the state information about
the network with much greater accuracy than the
existing algorithms. The new approach achieves very
good performance in terms of delay deviation,
success ratio and crank back ratio
3.1. Peformance Analysis and Simulation Results
LSRA is compared to a BP (Best point), WP
(worst point), & KK (modified Korkmaz Krunz)
algorithm.
WP: The worst delay and bandwidth
parameter is used to represent a mesh link.
BP: The best delay and bandwidth
parameter is used to represent a link.
KK: A curve is used to approximate the
points on a delay-bandwidth plane.Each link is
represesnted by delay, bandwidth and stretch factors
[9].
A feasible request may be rejected due to an
inaccurate approximation of dreqw. (Given the
bandwidth requirement reqw, the best delay value
supported by a line segment l is the corresponding
delay coordinate (dreqw)). We count the paths of a
request only when all four algorithms accept that
request.
The comparison metrics are:
Delay Deviation:
Delay Deviation measures the difference between the
real delay and the estimated delay obtained by the
aggregation.It is defined to be

Delay deviation = Estimated Delay of the shortest

path - Actual Delay of the shortest path.

Success ratio:
Success ratio is used to measure quantitatively how
well an algorithm finds feasible paths and it is
defined as Fig.7 Crank back ratio Vs. Bandwidth

Success ratio = Total no. of feasible paths found

Total no. of feasible requests

Crank-back ratio:
Crank back ratio measures how often crank back
happens and it is defined as

Crank-back ratio= Total no. of requests crank-backed

Total no.of requests accepted in
Inter- domain routing
Fig.8 Success Ratio -Crank back ratio Vs.
Bandwidth
LSRA has smaller delay deviations and a high
4. CONCLUSION
success ratio, a small crankback ratio, than other
algorithms. These are some of the simulation results
Thus this paper provides a novel QoS
as follows:
representation for topology aggregation in delay-
bandwidth sensitive networks. Line segments are
used in the delay-bandwidth plane instead of points
to represent the QoS parameters of logical links.
The paper discusses algorithms to compute line
segments for logical links of a mesh, to aggregate
links into a star with bypasses representation and to
find QoS routes using line segments

While there may be some low-end routers

in the domain that do not have the CPU/Memory
capacity to handle the aggregation, our approach
Fig.5. Delay Deviation of different aggregation does not require every node to participate in the
Schemes aggregation computation or to store the aggregated
topology. This flexibility helps to prevent
performance degradation of the critical border nodes
due to excessive number of interdomain routing
requests.

REFERENCES
[1]Private Network –Network Interface
Specification Version 1.0.Mar 1996.
[2] Y.Rekhter and T.LI, ”A border Gateway routing
Fig.6. Success ratio Vs. Bandwidth in networks with inaccurate information:heory and
algorithms”,IEEE/ACMTrans.Networking,vol.7.
pp.350-364,June 1999
[3 ]Roch A. Gu´erin and Ariel Orda “QoS Routing in
Networks with Inaccurate Information: Theory and
Algorithms”, IEEE/ACM Trans.
Networking,vol.7.pp.350-364,June 1999.
[4]F. Hao and E. W. Zegura, “On scalable QOS
routing: Performance evaluation of topology
aggregation”, INFOCOM, pp. 147–156, 2000.
[5]B.Awerbuch and Y.Shavitt,”Topology
aggregation for directed graphs” IEEE/ACM
Trans.Networking, vol.9.pp.82-90, feb.2001
[6]W.Lee,”Spanning tree method for link state
aggregation in large communication networks”, in
proc.IEEE INFOCOM, 1995, pp, 297-310.
[7]Atsushi Iwata and Hiroshi Suzuki “QOS
Aggregation Algorithms in Hierarchical ATM
Networks”, 1998
[8]T.Korkmaz and Krunz,”source oriented topology
aggregation with multiple QOS parameters in
hierarchical networks,” ACM Trans. Modeling
Comput. Simulation,vol.10.no.4.295-325.Oct.2000
[9] S. Chen and K. Nahrstedt, “An Overview of
Quality of Service Routing for Next- Generation
CERTIFICATE
This is to certify that the paper “Performance of Topology Aggregation Schemes in
Hierarchical Routing” is not published in any Journal’s, International conference and National Conference
anywhere.
Authors
1. A. Sherly Alphonse.
2. Mr. E. Baburaj.
High-Speed Networks: Problems and Solutions,”
IEEE Network Magazine, pp. 64–79,
November/December 1998.
[10]Zheng Wang and Jon crowcroft,”Bandwidth
Delay Based Routing Algorithms” in proc.IEEE
GLOBECOM vol.3 1995 .pp 2129
[11] Daniel Bauer, John N. Daigle, Ilias Iliadis and
Paolo Scotton “Efficient Frontier Formulation for
Additive and Restrictive Metrics in Hierarchical
Routing” in IEEE conf.communications
conf.Rec.vol.3,June 2000 ,pp.1353
[12] Z.Wang and J.Crowcroft,”Quality of service
routing for supporting multimedia applications
“IEEE J.Selected areas of communication,
pp.1228-1234, sept1996
[13]M.Faloutus, P.Faloutus, C.faloutus,”On power
law relationships of the internet topology”,in Proc.
ACM SIGCOMM, 1999 pp.251-262
[14]Jared Winick, Sugih Jamin,”Inet 3.0, Internet
Topology Generator,” University off Michigan
Technical Report”CSE-456-02, 2000
[15]B.M.Waxman,”Routing of multipoint
connections”IEEE.J.Select.areas on communication
“vol.6.pp.1617-1622.Dec/1988