Professional Documents
Culture Documents
i
i
B W = is the set of border nodes
and } {
ij
km
e E = ( j i )is the set of physical inter-domain
link (e
km
ij
) is the inter-domain link connects the border node k
in domain i and the border node m in domain j . Either the
intra-domain or the physical inter-domain link carries a
wavelength available vector in which the various wavelengths
are marked by 1or 0, which means whether the wavelength is
available or not. The graph ) , ( E W H does not have the full
connectivity of the network and the link cost depends on
abstraction strategy which is to be discussed below.
Simple Node (SN): this method condenses a domain to a
simple virtual node hiding all physical intra-domain links [12].
For example, as it show in the Fig. 2. The three domains in
Fig. 1 collapsed to three single nodes v1, v2, v3. All the
intra-domain links are not visible and just remain the physical
inter-domain links between the v1, v2, v3 and the costs of
these links are just ignored. This is the simplest abstraction
method meanwhile would have the worst routing
performance.
Full mesh abstraction (FM): this method preserve more
inside domain details than the simple node do [12]. It uses the
border nodes and the virtual links connecting them to
represent an abstracted virtual topology. The scheme will run
K-shortest path for each border node pairs (k,m) in domain i
to generate a set of paths } {
ij
km
i
e E = (i=j). Each path e
km
ij
between k and m maintain an available wavelength vector
which is gathered from the wavelength available vector in
every physical link along this path. The number of available
wavelengths C(e
km
ij
) in each e
km
ij
is counted. The path with
)) ( max(
ij
km
e C would be chosen as the intra-domain virtual
link between node pair (k,m). And the number of available
wavelengths of an inter-domain physical link is marked with
C(e
km
ij
) (ij) as the link cost. This abstraction strategy is
shown in Fig. 2. It shows a significant higher abstraction
complexity than the simple node method but would lead to a
better routing performance.
Hybrid abstraction (TH): The simple node scheme
provides no visibility into domain and has low abstraction
accuracy. Meanwhile, the full-mesh scheme provides more
accurate description of inner domain topology, albeit at the
cost of significant computational complexity. The hybrid
abstraction method is derived from the above two methods
[13]. To be specific, a domain will be represented by FM
abstraction, if the number of its intra domain links is less than
(b(b1)); otherwise, SS is abstraction is adopted as it shows
in Fig. 2. The computation complexity is between
( ) E O and ( ) ) (
2
+
i
i
n E O .
1
1
V
3
1
V 3
2
V
2
1
V
2
2
V
2
3
V
1
1
V
3
1
V
3
2
V
2
1
V
1
V
2
V
3
V
Figure 2. Topology abstraction
2. Inter-domain routing strategy
The multi-domain network supports the hierarchical
routing to block the information interchange among different
domains. Generally, three or above levels model will
significantly undermine the routing performance, so it is more
practical to adopt the two level routing model. The
inter-domain routing procedure in multi-domain network is
more complex than the single domain case, because the route
selection must rely on the topology of each routing level.
However, in the two level model, the routing procedure could
be simplified and divided into two steps: First, the candidate
LR (Loose route) is computed according to the high level
abstracted topology. Second, the LR will be expanded by the
ingress border node according to the low level physical
topology to get the ER (Explicit route).
The LR is computed using K-shortest path algorithm
based on the abstracted virtual topology. When the calculation
is finished, a set of LR sequences between the source and
destination called candidate LRs will be generated and later a
optimum LR among them will be further selected as the final
result. Traditionally, two similar selection schemes are mostly
used to select the final LR. One is minimum hop, the other is
minimum converters. However, both of them cant provide
best blocking performance for inter-domain lightpath. As a
result, more advanced selection scheme is needed, which is
discussed below:
Considering the wavelength constraint, there must be at
least one available wavelength in all links along the LR.
Assuming that the wavelength states in each hop are mutually
independent and the numbers of available wavelengths in the
adjacent hops are i and j, then the probability that the two
hops share n common wavelength is:
( )
j s
s
i
s
j
i s
C
C C
n
P P B
j
s
n j
i s
n
i
= =
=
0 0
2
(1)
Here S is the total number of wavelengths, P
s-i
is the
probability that the first hop has i available wavelengths
and P
s-i
is the probability that the second hop has j ava
ilable wavelengths. Moreover, the probability that the l h
ops link share n common wavelengths is:
( ) ( )
j s
l
i
s
i
s
j
C
C C l
n
P B B
j
s
n j
i s
n
i
= =
=
1
0 0
(2)
The probability that no common wavelength is shared
by the l hops is:
( ) ( )
j s
l
i
s
i
s
j
C
C l
P B B
j
s
j
i s
= =
=
1
0 0
0
(3)
B
0
(l)
also represents the blocking probability of the l hops link.
According to the recurrent formula shown in (3), the
blocking probability is not only decided by the hops l, but also
by the probability of the number of available wavelength that
a link possesses, in other words, the extent to which a link is
leisure or idle.
In this case, MLP (most leisure path) is proposed to take
both the hops factor and the available wavelength factor into
consideration. First, the MLP routing algorithm will find out a
sequence of candidate LR by using K-shortest paths. Then,
each LR will inspect its own links to find out
the )} ( { min
ij
km
LR e
e C
i
km
\
|
= =
n
s
n
n
s s
n
N S
w
w
w
w
w
w
w
w
w
W W W W
2
1
2
2
2
2
1
1
1
2
1
1
2 1
) , , , ( (5)
(6)
T
n N
d d d D ) (
2 1 1
=
= =
l
l N N S T
s S
D
D D W
w w w W
0
' ' ' '
1
2 1 1
(6)
ID-MU, the wavelength utilization rank array is generated
from the wave array. The available wavelengths in the wave
array will be sorted according to their utilization frequency in
descending order and each of them will attain a rank that
equals to their position in the sorted array. For example, the
most used wavelength will be given the rank number 1, the
second will be given 2 and so on. Besides, every domain
along the lightpath will be ranked in an identical manner to
generate the blocking probability rank array.
Then the inter-domain wavelength utilization rank array
will be calculated by the equation (6), where a new factor
i
w' called inter-domain wavelength utilization rank is defined
to represent the assignment priority of the available
wavelengths. The ID-MU will choose the wavelength with
minimum rank among the available wavelengths.
According to equation (6) , busy domains will be given
precedence over other domains when assigning the
wavelength to reduce their blocking probability. If D
l
is empty,
which means that the lightpath contains only an inter-domain
link where the inter-domain wavelength utilization rank array
is zero and all the available wavelengths have the same
priority.
Finally, the complete process of ID-MU is presented
below: First, the selected light path L is divided into N+1
sub-path
m
sub
L (1<m<N+1) by N converters which are
actually used. In transparent network, as there is no converter
(N=0), the light path will not be divided. Later, each available
wavelength vector belongs to virtual links and physical
inter-domain links along the lightpath will perform a logical
ADD operation to generate a new available wavelength vector
for the entire lightpath and
1
'
S
W should be calculated.
Then, a wavelength with maximal
i
w' in
1
'
S
W will be
assigned to the lightpath. In the translucent network, the
logical ADD operation is performed in each sub-path
m
sub
L ,
and the
N S
W
' is computed respectively in these sub-paths.
Then, each sub-path selects its own wavelength according to
1
'
S
W . Finally, the converters will work, if the adjacent
sub-link selected different wavelength.
Compared with the Random and the FF algorithm, this new
wavelength assignment algorithm has several advantages. First,
it can prevent the deterioration of block situation in the busy
domains and make full use of its wavelength resources.
Second, the average blocking probability in a network is lower
by using ID-MU than Random or FF. Third, the blocking
probability of inter-domain lightpaths is lower as well. Fig. 4
shows the ID-MU algorithm in translucent network, where the
lightpath is divided into two sub-path and each one chooses its
own wavelength according to
1
'
S
W .
Wave Rank Rank'
1 1 1
2
2 2
3
3 3
4
4 4
5
5 5
6
6 6
7
7 7
8
8 8
B
1
B
2
B
3
Most used wavelength is 1 Most used wavelength is 2
Lsub
1
assigned with 2
Lsub
2
assigned with 3
Most used wavelength is 3
3 has the minimum rank in W' Convert from 2 to 3
Wave Rank'
D1 D2
1 1 4 6
2 2 1 5
3
3 2 8
4
4 3 11
5
5 5 15
6
6 6 18
7
7 7 21
8
8 8 24
Rank Dom Rank
D1
2
D2
1
The wavelength utilization rank
matrix of domain-1 and
domain-2 and the inter-domain
wavelength utilization rank
array W'
2 has the minimum rank in W'
The blocking
probability rank
array of domain-1
and domain-2
The wavelength utilization
rank array of domain-3 and
the inter-domain
wavelength utilization rank
array W'
D1 D2
D3
Figure 4. ID-MU algorithm in translucent network
4. Inter-domain light path setup procedure
In the previous section we have discussed several key
steps in inter-domain lightpath setup procedure, including
topology abstraction, routing and wavelength assignment
algorithm. Now, the entire inter-domain light path setup
procedure will be demonstrated below.
First of all, the inter-domain lightpath provisioning in
multi-domain networks adopts RSVP-TE [22] signaling
system which is designed for resources reservation in
circuit-switching networks. Meanwhile, the topology
abstraction scheme is hybrid scheme to make a trade-off
between the computation complexity and the abstraction
accuracy. When an interior OXC node originates an
inter-domain light path setup requirement, this node first
sends a query message to the nearest border node. Then, a set
of loose route (LR) sequences will be computed according to
virtual topology and a candidate LR is chosen using certain
scheme such as minimum hops and proposed MLP. This LR
sequence involves the egress border node in source domain,
all egress and ingress border nodes in intermediate domains,
final ingress border in destination domain. If the source node
is a border node, some of the above step will only be
performed logically.
Next, the LR is packaged into a RSVP-TE PATH signaling
and sent downstream for further route expansion. Here ingress
border nodes in intermediate domains carry out the explicit
route expansion to resolve the link to the egress border nodes
in their domain. And the ingress border nodes in the
destination domain work out the explicit route to the
destination OXC node. If the path is fully expanded, a
wavelength is chosen by particular wavelength assignment
algorithm. Here ID-MU algorithm is recommended to solve
the assignment problem in transparent or translucent network.
Finally, the wavelength information is packaged into an
upstream RSVP-TE RESV message and sent back to the
source OXC node. All OXC nodes receiving a RESV message
will check and reserve the required wavelength on each link
and terminate setup if the wavelength is unavailable. The
procedure is shown in Fig. 5.
Figure 5. Flow chart of inter-domain lightpath setup procedure
IV. SIMULATION AND ANALYSIS
The testing network, which is built on NS2 simulation
software, contains seven domains with fifteen pairs of
unidirectional inter-domain links and fifty eight pairs of
intra-domain link. The number of wavelengths per fiber is set
to be sixteen .The average amount of inter-domain links is
4.28, which shows a good inter-domain connectivity. The
average amount of intra-domain links is 16.6. To ensure the
blocking probability varies from domain to domain, the traffic
load in each domain is different. First the simulation program
tests the performance of the network only with intra-domain
connection requirement. Then the intra/inter-domain
requirements are launched simultaneously at a ratio of 5/5 to
test the inter-domain performance. The blocking probability
including the inter, intra, and average will be measured
together. Total requirements are average over 10
5
attempts and
the traffic load is increasing from 0 to 300 Erlang.
In Fig. 7, different kinds of topology abstraction, Simple
node, Full mesh and Hybrid are compared. The routing and
wavelength assignment strategies are fixed to be MLP and
ID-MU. The result in Fig. 7 shows that Simple node method
has the worst performance among the tree strategies, in
contrast, the Full mesh method shows the lowest blocking
probability. Because the Full mesh abstraction can provide
more accurate topology to the routing strategy to gain better
routing performance. The performance of Hybrid method is
better than the simple node method but worse than Full mesh
method as expected, for it makes a trade-off between the
routing accuracy and the calculation complexity or routing
overhead. When the traffic load is light, the performances of
the three methods are very similar to each other. However,
when the traffic load grows, the performance varies a lot.
In Fig. 8, the proposed MPL routing strategy is compared
with the minimum hop count strategy. The topology
abstraction is supposed to be Hybrid abstraction. The ID-MU
is chosen as the wavelength assignment strategy. The curves
in Fig. 8 show that the MLP gets better performance than the
minimum hop only when the traffic load is heavy. Because
when the load is light, both the long paths and short paths
have enough available wavelength resources and the MLP
prefers to choose short path just like the minimum hop does.
However, when the load grows, different from the minimum
hop strategy, the MLP will give up some short but busy path
to avoid block.
Figure 6. Simulation network
Figure 7. Inter-domain Bp with different topology abstraction scheme
Figure 8. Inter-domain Bp with different routing strategy
Finally, two wavelength assignment algorithms are tested,
which are the proposed ID-MU algorithm and FF algorithm.
Fig. 9 demonstrates the blocking probabilities in the domain
with heavy traffic load ranging from 120 to 300 Erlang with
and without inter-domain lightpath requirement. As it shows,
blocking probability grows after the inter-domain connection
requirement appeared. However, the growth is less significant
by using the ID-MU method than by FF, because the ID-MU
can optimize the wavelength utilization in domain with heavy
traffic, which is helpful to reduce the blocking probability.
Fig. 10 shows the average blocking probability and the
blocking probability of inter-domain lightpath in transparent
network by using different wavelength assignment strategy.
As it shows, with the same traffic load, the inter-domain
blocking probability is higher than the average, because the
number of inter-domain links is generally less than number of
intra-domain links, which makes the routing process harder.
However, in both cases, the ID-MU shows better performance
than the FF.
In translucent network, as it shows in Fig.11, the blocking
probability is lower because of the use of wavelength
converter, and the ID-MU also shows better performance.
Figure 9. Bp in domain with heavy traffic
Figure 10. Bp in transparent network
Figure 11 . Bp in translucent network
V. CONCLUSION
In this paper, the inter-domain RWA for distributed
DWDM network is discussed. In addressing the inter-domain
lightpath setup problem, a topology abstraction strategy
named Hybrid is adopted which is derived from the FM and
SS strategy. Besides, considering both the hops and the
wavelength resources factor, a new inter-domain routing
strategy called MLP is proposed. In addition, to reduce
blocking probability, a new inter-domain wavelength
assignment algorithm called ID-MU is proposed as well,
which extends the MU algorithm to the inter-domain scenario.
The result shows that the Hybrid abstraction successfully
made a trade-off between routing overhead and the abstraction
accuracy and the MLP showed a better performance than the
previously used minimum hops method when the traffic load
grew. Finally, the ID-MU can effectively reduce both the intra
and inter domain blocking probability.
REFERENCE
[1] N. Ghani. J. Pan, X. Cheng, Metropolitan optical networks, Optical
Fiber Telecommunications IV, Academic Press, London, March 2002,
pp. 329403 (Chapter 8)
[2] G. Bernstein, B. Rajagopalan, D. Saha, Optical Network
Control-Architecture, Protocols and Standards, Addison Wesley,
Boston, 2003.
[3] ITU-T Rec.G8080/Y.1304, Architecture for the Automatically
Switched Optical Network(ASON), Novemeber 2001.
[4] J.Vasseur, Inter-Area and Inter-AS MPLS Traffic Engineering, IETF
Draft draft-vasseur-ccamp-inter-area-as-te-00.txt, 2004.
[5] B. St. Arnaud et al., BGP optical switches and lightpath route arbiter,
Optical Networks Magazine 2 (2), pp.7381, 2001.
[6] W. Alanqar, A. Jukan, Extending end-to-end optical service
provisioning and restoration in carrier networks: opportunities, issues,
and challenges, IEEE Communications Magazine, pp.5260, 2004.
[7] I.Iliadis, Optimal PNNI complex node representations for restrictive
costs and minimal path computation time, IEEE/ACM Transactions on
Networking 8 (4), 2000.
[8] X. Yang, B. Ramamurthy, Inter-Domain Dynamic Routing in
Multi-Layer Optical Transport Networks, IEEE GLOBECOM 2003,
San Francisco, CA, December 2003.
[9] S. Sanchez-Lopez, et al, A Hierarchical Routing Approach for
GMPLS-Based Control Plane for ASON, IEEE ICC 2005, Seoul,
Korea, June 2005.
[10] P. Ho, H.T. Mouftah, A novel survivable routing algorithm for
segment shared protection in mesh WDM networks with partial
wavelength conversion, IEEE Journal of Selected Areas on
Communications 22 (8), pp.15391548, 2004.
[11] Q. Liu et al., Hierarchical Inter-Domain Routing in Optical DWDM
Networks, IEEE INFOCOM 2006 High-Speed Networking Workshop,
Barcelona, Spain, April 2006.
[12] T. Korkmaz and M. Krunz, Source oriented Topology Aggregation
with Multiple QoS Parameters in Hierarchical Networks, IEEE/ACM
Transactions on Modeling and Computer Simulation, vol. 10, pp.
295325, Oct. 2000.
[13] Guido Maier, Chiara Busca, Achille Pattavina, Multi-Domain Routing
Techniques with Topology Aggregation in ASON Networks, 2008
International Conference on Optical Network Design and Modeling
(ONDM), pp.1-6, March 2008.
[14] I. Chlamtac, A. Ganz, and G. Karmi, Purely Optical Networks for
Terabit Communication, Proc., IEEE INFOCOM 89,Washington, DC,
vol. 3, pp. 887-896, April 1989.
[15] R. A. Barry and S. Subramaniam. The MAX-SUM Wavelength
Assignment Algorithm for WDM Ring Networks, Proc., OFC 97, Feb.
1997.
[16] A. Birman and A. Kershenbaum, Routing and Wavelength Assignment
Methods in Single-Hop All-Optical Networks with Blocking, Proc.,
IEEE INFOCOM 95, Boston, MA, vol. 2, pp. 431-438, April 1995.
[17] G. Jeong and E. Ayanoglu, Comparison of Wavelength-Interchanging
and Wavelength-Selective Cross-Connects in Multiwavelength
All-Optical Networks, Proc., IEEE INFOCOM 96, San Francisco, CA,
vol. 1, pp. 156-163, March 1996.
[18] E. Karasan and E. Ayanoglu, Effects of Wavelength Routing and
Selection Algorithms on Wavelength Conversion Gain in WDM
Optical Networks, IEEE/ACM Transactions on Networking, vol. 6, no.
2, pp. 186-196, April 1998.
[19] S. Subramaniam and R. A. Barry, Wavelength Assignment in Fixed
Routing WDM Networks, Proc., ICC 97, Montreal, Canada, vol. 1, pp.
406-410, June 1997.
[20] X. Zhang and C. Qiao, Wavelength Assignment for Dynamic Traffic in
Multi-fiber WDM Networks, Proc., 7th International Conference on
Computer Communications and Networks, Lafayette, LA, pp. 479-485,
Oct. 1998.
[21] H. Zang, J. Jue, B. Mukherjee, A review of routing and wavelength
assignment approaches for wavelength- routed optical WDM networks,
Optical Networks Magazine 1 (1), 2000.
[22] The Multiprotocol Label Switching (MPLS) Working Group decision
on MPLS signaling protocols, RFC3468, L. Andersson and G. Swallow,
February 2003