OPTIMIZATION OF IP NETWORKS IN VARIOUS HYBRID IGP/MPLS ROUTING SCHEMES

12th GI/ITG CONFERENCE ON MEASURING, MODELLING AND
EVALUATION OF COMPUTER AND COMMUNICATION SYSTEMS
3rd POLISH-GERMAN TELETRAFFIC SYMPOSIUM
OPTIMIZATION OF IP NETWORKS IN VARIOUS HYBRID
IGP/MPLS ROUTING SCHEMES
Eueung Mulyana, Ulrich Killat
Department of Communication Networks, Hamburg University of Technology (TUHH)
Address: BA IIA, Denickestrasse 17, 21071 Hamburg, Germany
E-mail: {mulyana,killat}@tu-harburg.de
Abstract
This paper presents an offline traffic engineering (TE) approach in several hybrid IGP/MPLS
environments. Though IGP (Interior Gateway Protocol) routing has proven its scalability and
reliability, effective traffic engineering has been difficult to achieve in public IP networks because of
the limited functional capabilities of conventional IP technologies. MPLS (Multi-Protocol Label
Switching) on one hand enhances the possibility to engineer traffic on IP networks by allowing
explicit routes, but on the other hand it suffers from the scalability problem. Hybrid IGP/MPLS
approaches rely on IP native routing as much as possible and use MPLS only if necessary. In this
work we propose a novel hybrid traffic engineering method based on genetic algorithms and
investigate it in several IGP/MPLS routing schemes. This can be considered as an offline TE
approach to handle long or medium-term traffic variations. In our approach the maximum number of
hops as well as the maximum delay an LSP (Label Switched Path) may take and the maximum
number of LSPs that can be installed in the network are treated as constraints. We apply our method
to the German scientific network (G-WiN) for which a traffic matrix was randomly generated. We
will show results for several hybrid routing schemes and compare them to the result of pure IGP
routing.
Keywords
routing, traffic engineering, metaheuristics, evolutionary computation, IP networks, MPLS
1. INTRODUCTION
As the candidate for the future multi-service communication platform, IP
technology has to support a wide range of quality of service assertions as required
by different types of applications. In this respect, it should also be able to provide
the necessary management flexibility for controlling traffic distribution over the
network in order to keep network resources within desired operational conditions,

which are necessary for guaranteeing quality of service. Hence,traffic engineering
(TE), i.e. mapping traffic flows onto the existing physical network topology in the
most effective way to accomplish desired operational objectives [3] has become an
important issue in IP networks. There are several approaches for deploying TE in
current IP networks e.g. by optimizing the parameters used for routing decisions,
so that a better network performance will be obtained [4-10,13,16,17], or by using
explicit routing in an overlay model with ATM or Frame Relay technology. Recent
developments in Multiprotocol Label Switching (MPLS) open new possibilities to
address some of the limitations of IP systems concerning traffic engineering. In
particular MPLS efficiently supports origin connection control through explicit
label-switched paths (LSPs). In an MPLS network, it is possible to explicitly
specify one or several paths for each traffic demand from a source to a destination.
By using a full mesh of LSPs, the traffic matrix of source to destination flows in a
network can easily be obtained. Because of scalability issues in a full mesh
architecture and for seamless migration from the current IP network running IGP
(Interior Gateway Protocol), the ISPs may adopt a tactical approach to MPLS, in
which they create LSP-tunnels only when necessary, for example to address
specific congestion problems. Although this approach does not fully profit from the
benefits of MPLS, it is an attractive alternative compared to the traditional TE
method. In the context of pure MPLS networks, this approach can be thought of as
a mixed TE approach using both connectionless and connection-oriented routing
capability of MPLS routers. To the best of our knowledge, there are only a few
publications that consider IGP/MPLS scenarios for offline traffic engineering
[5,11,14,15,18]. In [5] three different models are presented. In the first model
(basic IGP shortcut) a packet will be forwarded to an LSP if its destination is the
tail-end of the LSP. In the second model (IGP shortcut) all packets to nodes that are
the tail-ends of LSPs and to nodes that are downstream of the tail-end nodes will
flow over those LSPs. In the last model LSPs are advertised in the IGP and used in
the shortest path calculation as virtual interfaces. In these three models IGP and
MPLS are working together in the same layer i.e. IGP routing is modified taking
into account LSPs. Recent work such as [11,14] presents an overlay model where
IGP and MPLS are working separately. In our investigations we consider three
hybrid routing schemes, which will be discussed in detail in the next Section. In the
following we first formulate the problem and introduce some notations. In Section
3 we present the genetic algorithm for solving the problem. After that in Section 4
we present some results and analysis for the network instance i.e. the German
scientific network (G-WiN) from [1] for which the traffic matrix was randomly
generated.

Optimization of IP Networks in Various Hybrid IGP/MPLS Routing Schemes
2. PROBLEM FORMULATION
Now we will formulate the problem in mathematical notation. A directed
network ),( ANG = is given, where N is the set of nodes representing the
network’s routers and A is the set of arcs representing the network’s links. Each
link Aji ∈),( has a capacity jic , . Furthermore, we have a demand vu
f ,
for each
pair NNvu ×∈),( , giving the demand to be carried from source u to destination v.
A set of LSPs is denoted by Π and indexed by k. An kLSP consists of a loop-free
node sequence ),,( kk th L where kh , kt denote the head and tail node,
respectively. A real variable vu
jil ,
, is associated with the load on link ),( ji resulting
from flow demand vu
f ,
along shortest path routing, and k
jilLSP
, resulting from the
flow or the flow aggregate in kLSP ( k
f LSP
). Note that for simplicity, in this paper
we do not consider ECMP (Equal Cost Multi-Path) in case that several shortest
paths exist. It means that the ECMP feature is either disabled or using optimized
metrics that result in a unique shortest path routing pattern [4,6,7,17].
Figure 1 Shortest path trees for node 1 in three different scenarios : (a) Basic IGP shortcut,
(b) IGP shortcut, (c) Overlay
Consider the network in Figure 1 with an LSP originating from node 2 and
ending at node 7 (via nodes 4-6-8). In IGP/MPLS basic shortcut (BIS) scenario all
packets arriving in node 2 with destination of node 7, will be forwarded to the LSP.
Figure 1 (a) shows the shortest path tree for node 1. It is obvious that in the BIS
model only the node 7 is reachable via the LSP, while the other flows will follow
the normal path. In the IGP shortcut (IS) scenario all packets arriving in node 2
with destination of node 7 as well as of its downstreams i.e. the nodes 8 and 9 will
be forwarded to the LSP. Figure 1 (b) shows the shortest path tree for node 1 in the
IS model. Note that the link (5,7) will not be used for routing all traffic originating
from node 1 anymore. Figure 1 (c) shows the shortest path tree for node 1 in
overlay (OV) model. In this model an LSP is used only to route traffic from its
4 5
8 9
6
1
5
3
2
6
7 8
9
4
(b)
7
LSP
8 9
(a)
7
LSP
(c)
7
LSP
1
1
1
1
1
1
2
2
3
2
2
1
LSP
1
2 3
4 5
6
1
2 3
4 5
8 9
6
1
2 3
4 5
8 9
6
1
5
3
2
6
7 8
9
4
(b)
7
LSP
8 9
(a)
7
LSP
(c)
7
LSP
1
1
1
1
1
1
2
2
3
2
2
1
LSP
1
2 3
4 5
6
1
2 3
4 5
8 9
6
1
2 3

$ 1 452 6 $ 6
$ * $ $ $ )1 # @ ( ! "
DE $ $ ( 4521 4
"/!".! (*( === − αααα $ $
Π Π 452
α $ $ α > $Π Π
$ !$ $ " $
$ 6 $Π 1
$ "! $ %
+= 452
!("
DE $ %
∀≠∈
=
+
"!"!$"
!*"
=
=
+
452"!$"!"!452
!;"
&'5 $ '5 $ %
∈∅≠∈∅=
=
+
$HI$H "" "
"
"
"
!>"
=
+
452"!452
452
!B"
452
$ $ 452 1 1 452
≥
= $ = 1 @ 6 ×∈∀= "!"! $
&' ∈∀= "!"! $ 452
$ %
J/J. 6( Π+⋅ ρ !?"
∈∀≤ "!6ρρ !)"
=ρ = "! 1 # ( $
$ 1 7 !?" 6ρ
=$ $ JJ Π
452 $ $ 6 1
9 6ρ $ 1 '
@ *
-* !@ * "1 $ 6 $

all such links for computing maxρ in (7). Furthermore, it is important to limit the
number of hops and the delay for the LSPs:
khN k ∀+≤ ,1|| maxLSP (8)
kk ∀≤ ,maxLSP δδ (9)
where kNLSP denotes the set of nodes that belong to the kLSP , kLSPδ the delay
introduced by kLSP , maxh the maximum allowable hop-count and maxδ the
maximum allowable delay. Having the traffic matrix, the metric values and a set of
LSPs, we can compute the load distribution on the network. Every solution has a
quality measure according to (6). Although a solution is feasible if 1, ≤jiρ or
correspondingly 1max ≤ρ , the optimization is performed with no constraints to
force this condition, but we simply minimize the objective function. The desired
result is a set of LSPs which corresponds to the minimized cost function and to the
certain performance parameters (e.g. hop-count, delay). Although here we treated
the set W as a given set, the method presented can easily be integrated in a metric
based optimization approach to address combined problems, for example: some
LSPs are created when the metric based approach fails to further improve network
performance or vice versa. Since the equations (1)-(5) depend on the set Π , which
actually has to be determined, the formulation for IGP/MPLS routing presented
here might only be suitable for the heuristic solving method to be addressed in
Section 3.
Figure 2 The G-WiN network taken from [1]
3. A GENETIC ALGORITHM FOR HYBRID IGP/MPLS TE
Genetic algorithm (GA) is a population-based search method, that is adopted
from nature. The population consists of individuals or chromosomes that represent
solutions to the problem. Hence, the first design challenge of the GA is how to
2
1
5
3
4
6
7
9
10
11
12 13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2
1
5
3
4
6
7
8
9
10
G-WiN Level 1
G-WiN Level 2
8
2
1
5
3
4
6
7
9
10
11
12 13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2
1
5
3
4
6
7
8
9
10
2
1
5
3
4
6
7
8
9
10
G-WiN Level 1
G-WiN Level 2
8

encode a solution in terms of a chromosome. The next step is to use this encoding
method to produce an initial population by randomly generating a suitable number
of chromosomes. There are no standardized rules to decide how many
chromosomes should be in the population. A size of around 30 up to 200
chromosomes is typically enough, because the number of chromosomes in the
population is not directly correlated with the quality of the solutions. After
generating this intial population, all chromosomes enter the evolution loop,
consisting of selection and some processes to form new chromosomes. At the
beginning of each iteration some vectors of high quality are selected as parent
chromosomes, which by applying the genetic mechanisms ”crossover” and
”mutation” will hopefully produce some better solutions for the next generation.
The least successful chromosomes of the previous iteration will be removed and
then be substituted by the new ones. Applying the described processes in many
iterations we continuously improve the average quality of the solution vectors until
the termination condition is satisfied. The GA used for our hybrid IGP/MPLS
problem, is basically a slight modification from that in [13]. Due to space reason,
we will simply refer to [13] for the selection, crossover and mutation algorithm and
in the following only discuss the encoding method, since it is the main difference
to the GA in [13].
Figure 3 Chromosome representation (for the case of asymmetric LSPs)
Encoding. In order to apply a genetic algorithm to the problems defined in Section
2, in general a suitable encoding of possible solutions in a vector (i.e. chromosome)
representation is needed. In our case a chromosome is represented by a set of
numbers 〉〈 Ll yyyy ,,,,, 21 KK where yl is an integer and ],0[ max
l
l cy ∈ . Each
position l is related to a certain node pair and the corresponding flow respectively.
Figure 3-(a) shows a simple case of the association of the flows to each position in
the chromosome. The set of flows or node pairs where an LSP may be installed
could simply be given or chosen on a random basis. If 0=ly , no LSP for the node
pair will be installed and the flow will be forwarded to the next hop node according
to the shortest path computation. If 0≠ly , an LSP for the node pair will be
installed and the flow will be routed to a that LSP. The set of possible LSPs for
val.
u=1 u=2 u=3 u=4
v=1
v=2
v=3
v=4
f(1,2) f(4,2)
f(1,3) f(2,3) f(4,3)
f(2,1) f(4,1)
f(3,2)
f(3,4)f(2,4)f(1,4)
f(3,1)
-
-
-
- 1
2
3
4
index LSP candidates
(a) (b)
f(3,1) f(4,2) f(1,3) f(2,3) f(4,3)
val. val. val. val.
f(1,2)
val.
The traffic matrix f(u,v)
chr. 0202
LSP D
LSP C
LSP B
LSP A LSP E
LSP F
LSP G
LSP I
LSP H LSP J
LSP K
LSP L
LSP M
LSP N
LSP O
LSP P
LSP Q
LSP R
1 3

flow vu
f ,
associated with the position l is obtained by applying k-shortest paths
algorithm with respect to the number of hops. This set will then be processed by
removing the LSP candidate, which is exactly the same as the original shortest path
and those which do not fulfill the constraints (8) and (9). The remaining LSPs are
put in a list and the value of yl points directly to the position of the LSP in the list.
It is not necessary to sort the LSP candidates according to some criteria (e.g.
number of hops) because changing the value of yl is performed on a stochastic basis
by the genetic operators (crossover and mutation). Figure 3-(b) shows an example
of the representation of a chromosome together with its interpretation. The
cardinality of the LSP candidates for each position (the constant l
cmax ) could be
different depending on the result of k-shortest paths algorithm or a given upper
bound :


 ≤
=
else
||if||
given
given
max
c
cPP
c
lll
(10)
_
where givenc is the given upper-bound and lP a set of all possible LSPs for the flow
associated with the position l. Using this representation we limit the number of
LSPs to be installed in the network by L.
Table 1 Some typical computation results (L=48)
4. RESULTS AND DISCUSSION
As mentioned in Section 1, for the following results we used the German
research and scientific network G-WiN network shown in Figure 2. It consists of
27 nodes (10 level-1 nodes and 17 level-2 nodes) and 76 links. Each level-1 link
has a transmission capacity of either 2.5 Gbps or 10 Gbps while each level-2 link
has either 6222× Mbps or 5.22× Gbps [1]. The traffic matrix is composed of

702 flows and each of them was generated randomly in the interval [4,355] Mbps.
The mean demand value is 34.6 Mbps and around 75% of the demands are below
the value of 30 Mbps. From the topology point of view, the optimization problem
for this network could be implemented in a hierarchical way i.e. we concentrate in
optimizing the level-1 network using the aggregated flows from all level-2 nodes.
In this way the complexity will be drastically reduced. For instance if we consider
the total number of LSPs for a pure MPLS network, with the hierarchical approach
we could reduce the total number of LSP’s by 87%, with the assumption that each
flow can not be split. However in order to keep the generality of the method in our
experiments we used a plain approach and considered both, level-1 and level-2
networks, in the same layer. The genetic algorithm was set to terminate if it
reached a value of 150 iterations or after 70 iterations with no more improvements.
Delays are modelled statically and consist of three components: the propagation
delay propδ , the processing and serialization delay spδ (link-rate dependent) and
delay margin cδ . The parameter setting was : constant 10001 =c (6), maximum
hop-count 4max =h (8), 12max =δ ms (9), 1sp =δ ms for 622 Mbps lines and 1c =δ
ms per link. In all experiments, the set of possible node pairs to be connected by an
LSP is given: an LSP may be installed for every two level-1 nodes, which are not
directly connected. Hence, the number of LSP can be installed in the network is
upper-bounded by L=48. The ECMP feature was disabled and the metric value was
originally set inversely proportional to the link’s capacity. Some typical results for
each hybrid scheme are displayed in Table 1.
Network Utilization. Traffic on level-2 links is not reroutable, thus for the
optimization they were marked as unconsidered and maxρ in (6) was substituted
with cons
maxρ i.e. the maximum utilization on the level-1 network. Using the original
IGP routing cons
maxρ is about 77.25% for the link (5,6). With optimization we can
reduce the value to 42.64% depending on the routing schemes and trade-off factors
(i.e. the number of LSPs to be installed). In general, a better utilization value will
be obtained by using more LSPs. The best result is obtained for IS model. This
could be the effect of the aggregation capability in the IS model. This aggregation
capability is indicated by the parameter LSP
ω which denotes the number of
different flows carried by an LSP. The average value of LSP
ω tends to increase for
the IS model.
Network Delay. Improving network utilization in all cases will increase the
average delay in the network, because to avoid hot-spot the flows often must
follow longer routes. By adding delay and hop constraints as in (8) and (9) we try
to trade between these performance parameters. Figure 4 shows the source-
destination delay before and after optimization, for the case of L=48 (Table 1).

Compared to the original IGP routing, the average delay at the end of the
optimization increases by the value of about 0.05 ms (OV), 0.21 ms (BIS) and 0.54
ms (IS) respectively. These differences could again be seen as a direct consequence
of each hybrid scheme. The number of demands, which are rerouted on longer
paths for each scheme, increases from 1.42% (OV) to 21.08% (IS). An almost
similar case happens for the routing hop-count: the average hop-count at the end of
the optimization increases, at most by the value of about 0.18 (IS). However, in
contrast to Figure 4 which indicates that some of demands are rerouted through
paths with less delay, there are no demands rerouted with less hop-count i.e. they
are rerouted either through paths with equal hop-count or through those with higher
hop-count.
Figure 4 Source destination delays for all schemes
5. SUMMARY AND CONCLUSION
In this paper we have considered the problem designing LSPs for hybrid
IGP/MPLS traffic engineering schemes and proposed a novel approach based on
genetic algorithms. Our results show that by constructing a few LSPs we could
improve network efficiency by reducing the maximum utilization in G-WiN
Average = 6.7377 ms
#flows (with delay increase)=1.42 %
#flows (decrease) = 0.28 %
#flows (similar) = 98.3%
#flows (similar) = 73.22 %
Average = 6.6874 ms
Average = 6.9038 ms
#flows (increase) = 5.7 %
#flows (decrease) = 1 %
Average = 7.2376 ms
Average = 6.7377 ms
#flows (with delay increase)=1.42 %
#flows (similar) = 98.3%
Average = 6.6874 ms
Average = 6.9038 ms
#flows (decrease) = 1 %
Average = 7.2376 ms

network scenario from 77% to 43% while simultaneously keeping the overall
network performance in terms of delays and hop-count within an acceptable range.
Although in IGP/MPLS schemes an ISP does not have all available features that
MPLS may offer, for example a source destination flow measurement, the
approach seems to be an attractive alternative and complementing the traditional
offline traffic engineering by optimizing IGP link metric.
BIBLIOGRAPHY
[1] Adler H. M., Neues im G-WiN, 37. DFN-Betriebstagung, November 2002.
[2] Awduche D., MPLS and Traffic Engineering in IP Networks, IEEE Communication
Magazine, December 1999.
[3] Awduche D., Chiu A., Elwalid A., Widjaja I., Xiao X., Overview and Principles of
Internet Traffic Engineering, RFC 3272, May 2002.
[4] Ben-Ameur W., Gourdin E. , Liau B. , Michel N., Determining Administrative Weigths
for Efficient Operational Single-Path Routing Management, Proceedings of 1st
PGTS,
2000.
[5] Ben-Ameur W., Michel N., Liau B., Geffard J., Gourdin E., Routing Strategies for IP
Networks, Telektronikk Magazine, 2/3, 2001.
[6] Bley A., Koch T., Integer Programming Approaches to Access and Backbone IP
Network Planning, Preprint ZIB ZR-02-41, 2002.
[7] Bourquia N., Ben-Ameur W., Gourdin E., Tolla P., Optimal Shortest Path Routing for
Internet Networks, Proceedings of 1st
INOC, 2003.
[8] Fortz B., Thorup M., Internet Traffic Engineering by Optimizing OSPF Weights,
Proceedings of IEEE Infocom, March 2000.
[9] Gajowniczek P., Pioro M., Szentesi A., Harmatos J., Solving an OSPF Routing Problem
with Simulated Allocation, Proceedings of 1st
PGTS, 2000.
[10] Karas P., Pioro M., Optimisation Problems Related to the Assignment of
Administrative Weights in the IP Networks’ Routing Protocols, Proceedings of 1st
PGTS,
2000.
[11] Koehler S., Binzenhoefer A., MPLS Traffic Engineering in OSPF Networks – A
Combined Approach, ITC 18, September 2003.
[12] Moy J., OSPF Version 2, RFC 2328, April 1998.
[13] Mulyana E., Killat U., A Hybrid Genetic Algorithm Approach for OSPF Weight
Setting Problem, Proceedings of 2nd
PGTS, 2002.
[14] Riedl A., Optimized Routing Adaptation in IP Networks Utilizing OSPF and MPLS,
Proceedings of IEEE ICC, May 2003.
[15] Shen N., Smit H., Calculating IGP Routes over TE Tunnels, Internet Draft, December
1999.
[16] Staehle D., Koehler S., Kohlhaas U., Optimization of IP Routing by Link Cost
Specification, Tech. Report, University of Wuerzburg, 2000.
[17] Thorup M., Roughan M., Avoiding Ties in Shortest Path First Routing, ___________.
[18] Wang Y., Zhang L., A Scalable Hybrid IP Network TE Approach, Internet Draft, June
2001.

OPTIMIZATION OF IP NETWORKS IN VARIOUS HYBRID IGP/MPLS ROUTING SCHEMES

More Related Content

What's hot (18)

Viewers also liked (11)

Similar to OPTIMIZATION OF IP NETWORKS IN VARIOUS HYBRID IGP/MPLS ROUTING SCHEMES (20)

Recently uploaded (20)

OPTIMIZATION OF IP NETWORKS IN VARIOUS HYBRID IGP/MPLS ROUTING SCHEMES