AN IMPROVED GMPLS SURVIVABILITY MECHANISM USING
LINK DELAY-CONSTRAINED ALGORITHM
Anastasios Bikos
2
, Christos Bouras
1,2
and Kostas Stamos
1,2,3
1
Research Academic Computer Technology Institute, Patras, Greece
2
Computer Engineering and Informatics Department, University of Patras, Patras, Greece
3
Technological Educational Institute of Patra, Patra, Greece
Keywords: GMPLS, Link failure probability, Link delay-constrained Algorithm, LSP routing, Path protection.
Abstract: It is widely accepted that GMPLS (Generalized MPLS) will be a key technology in the evolution of the next
generation of reliable Internet Protocol (IP) backbone networks. Conventional GMPLS-based optical-
switching network fault recovery only provides resiliency in terms of path segment selection instead of
constraint-based calculation. This can create severe impact on the protocol’s transport plane when a fault
occurs to a link or path with many optical connections attached to it. This paper proposes the
implementation of an improved GMPLS recovery algorithm based on the metric of optical link delay which
is achieved through the pre or post selection of a safer and more stable protection path with fewer
connections attached to it, and therefore with a lesser link delay metric compared to other possible paths.
The improved recovery algorithm is evaluated using the network simulator ns-2 and more particularly a
specialized simulator add-on for GMPLS, called ASONS (Automatically Switched Optical Network
Simulator). The results indicate improved resiliency, increased fault avoidance, and reduced packet loss.
1 INTRODUCTION
As IP traffic becomes more and more massive, the
optical switching network emerges as the most
promising solution for meeting the modern
backbone network needs. This has caused the
introduction of GMPLS (Generalized Multi-Protocol
Label Switching), as a reliable and stable optical
framework, in order to meet those new standards and
developments of telecom services (Dutta, 2008;
Stern, 2009; Farrel, 2006).
While optical fiber medium using wavelength
division multiplexing (WDM) offers tremendous
transmission bandwidth to deliver high-traffic
services cost effectively, faults such as the
unavailability of optical links are still an important
issue to resolve. Because the most massive amount
of traffic is transmitted over the optical backbone
network, a fault in the backbone may result in very
important service degradation. This forces Internet
Service Providers (ISPs) to include network
reliability parameters in their Service Level
Agreements and to design new protection strategies
guaranteeing fast failure recovery times and high
levels of reliability.
In this paper, we describe the design and
performance analysis of a QoS-Constrained GMPLS
Recovery algorithm, based on the ideas that have
been presented in (Ortega, 2004) for MPLS, which
is based on the dynamically changing optical link
Delay Constraint, distributed by the IGP Routing
Protocol. This makes the GMPLS fault recovery
procedure able to adapt to the current network state
and based on that condition it then becomes possible
to post compute and configure the backup path.
Finally, we evaluate the implementation of the
currently existing GMPLS Protection Mechanisms,
as well as an improved Restoration Mechanism,
based on this new algorithm, under the network
simulator ns-2 environment.
2 NETWORK SURVIVABILITY
ISSUES AND RECOVERY
SCHEMES
Next-generation optical communication technologies
(DWDM/OADM/PXC) are expected to exceed
aggregate capacities of hundreds of terabits per
45
Bikos A., Bouras C. and Stamos K..
AN IMPROVED GMPLS SURVIVABILITY MECHANISM USING LINK DELAY-CONSTRAINED ALGORITHM.
DOI: 10.5220/0003433900450050
In Proceedings of the International Conference on Data Communication Networking and Optical Communication System (DCNET-2011), pages 45-50
ISBN: 978-989-8425-69-0
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
second. As wavelength routing and all-optical
switching paves the way for network throughput of
such scales, network survivability assumes critical
importance. A single loss of or damage to a fiber is a
common means of a greater loss. A short network
outage can lead to huge data loss, particularly in the
backbone core. Thus, a connection being carried in
the network also needs high protection and
resilience. Survivability refers to the ability of the
network to reconfigure and re-establish
communication upon node and/or link failures.
Such network survivability can be classified into
two general categories: pre planned protection and
dynamic restoration. In pre-planned protection-based
techniques resources are already planned, typically
at the time of establishing a lightpath connection, to
recover from network failures and hence recovery is
faster.
During the normal operation phase these
reserved resources remain idle. Upon the occurrence
of failure, reserved resources are used to recover
from the failure according to protection protocols. In
contrast, in dynamic restoration, the resources used
for recovery from failure are not reserved at the time
of connection establishment, but are discovered
dynamically using link state algorithms when a
failure occurs. As it is obvious, dynamic restoration
uses resources efficiently, but the restoration time is
usually longer, because it requires the establishment
of a new functional backup path. Moreover, 100%
service recovery cannot be guaranteed as it is not
guaranteed that the spare capacity is available at the
time of failure (Dutta, 2008).
2.1 Problems with Conventional
GMPLS Restoration Mechanisms
One of the most common problems of the existing
fault recovery schemes in GMPLS networks is that
they do not consider the already existing link load of
a backup path when it has to be configured. A
typical bad case scenario is when selecting an
optical link which is a critical segment, or cut-edge,
for many connections. It has been shown that a
failure on this link has more overall impact on the
network traffic (Changwoo, 2007). Figure 1 shows a
related network situation where more connections
cross through a particular optical link, which
therefore acts as a bridge, than other links. As the
number of connections increases in a particular link,
so does the overall impact of a potential failure of
the link.
It is well known that some links have higher
failure probabilities, and this can be attributed to
their physical situation and conditions. This Link
Failure Probability Factor (LFP) is based on the type
of physical link, the node characteristics and
geographical distribution of the network segments.
Since these parameters are outside of our control, we
consider the LFP values for the network topology as
given, and our purpose is to route backup paths so
that traffic distribution across the network becomes
as even as possible. Thus we can try to decrease the
impact of new potential network faults in terms of
affected connections.
Figure 1: Link Failure in a Path with many connections
(Changwoo, 2007).
3 LINK DELAY-CONSTRAINED
ALGORITHM
IMPLEMENTATION
Due to the previous problems that occur from
traditional unconstrained Restoration Schemes in
GMPLS networks, our proposed algorithm
configures a backup path by searching for the
optimal path through the link state algorithm, based
on the delay parameter. The key concept of this
improved algorithm is that the path selection
procedure typically prefers links that carry fewer
connections, and thus, given the Link Failure
Probability Factors, distributes the impact of links
failures on LSP more evenly.
3.1 Constraint-based Algorithm
The implementation of our mechanism, from now on
called LDC (Link Delay-Constrained), is separated
in two phases: The Link Searching for Delay
Constraint Procedure and the Modified Dijkstra
Algorithm which takes into account the filtered link
information, with newer cost values, from the
previous phase according to the least delay
constraint. The delay metric can be calculated using
DCNET 2011 - International Conference on Data Communication Networking
46
on the fly measurements.
3.1.1 Link Searching for Delay Constraint
We present the following pseudo code that searches
for the optical link that satisfies the least link delay
constraint among all its neighbouring ones. In
particular, when it indeed finds this optimal link, it
sets all its nearby links, except itself, with higher arc
weight or cost (w(n, mi) >= 2), so that the Modified
Dijkstra’s Algorithm in the second phase will be
able to calculate the optimal path based on these new
link cost metrics, and thus re-update the whole
routing table. The optimal link selected from this
phase will continue to preserve the default cost value
(c=1), thus it will remain first selection priority for
the Dijkstra’s Algorithm, in order to create the
Protection Path.
L = All links except primary path links(l)
Function Link_Delay_Constraint_Search(L)
FOR each node n
Find link with min delay {lmin}
FOR each link l of n
IF l lmin
Set weight value of link l = 2
ELSE
lmin = 1 {default link cost}
Return L
Figure 2: Link Searching for Delay Constraint.
3.1.2 Modified Dijkstra’s Algorithm
The algorithm (Changwoo, 2007) receives the L
value from the previous phase and then searches and
selects a link to the destination t with the smallest
number of connections within pool L. Afterwards,
the optimal backup path is being configured. In any
case, even if an optimal path segment to a specific
node destination might not logically exist, due to the
previous link delay cost update, the Dijkstra’s
Algorithm (Changwoo, 2007) will compromise to a
path selection even with worse arc weight (w 2).
4 REDUCING THE IMPACT OF
THE LFP FACTOR WITH LDC
As discussed in the introductory section, in GMPLS-
based networks the usual method of recovering a
failure is the utilization of an alternative and disjoint
path to the main working path. The general time
process for the failure recovery procedure which
applies to both pre planned protection and dynamic
restoration is discussed in (Ortega, 2004).
Since the failure event point occurs, there is a
time period when data packets are inevitably being
lost due to the uncompleted switchover process, and
until the normalization procedure finishes. If the
survivability mechanism (either pre planned or
dynamic) does not consider the load of traffic
carried by each link of the potential protection LSP,
the restoration can become prone to more recurrent
faults and their associated costs. One such example,
where the impact of Link Failure Probability greatly
affects the future failure impact of the protection
LSP is shown and described in Figure 3.
Figure 3 (a): Link Failure Probability in the Working Path
with Conventional Resiliency.
Figure 3 (b): Link Failure Probability in the Working Path
with LDC Improved Resiliency.
Let as assume that the centralized optical links 3-
6 and 6-9 act as bridges for the overall network
traffic, thus contain a large number of optical
connections, as most traffic passes through them. In
Figure 3 (a) the working path (formed by the LSRs
1-3-6-9-11) contains the two links with high Link
Failure Probability (3-6 and 6-9).Unfortunately this
conventional method for path selection causes a high
risk of increased service impact, should a possible
link failure occur in the future. Indeed, when a
failure event occurs between the nodes 3 and 6 (one
of the two links mentioned earlier with high LFP
Factor), the GMPLS conventional survivability
AN IMPROVED GMPLS SURVIVABILITY MECHANISM USING LINK DELAY-CONSTRAINED ALGORITHM
47
procedure configures a segmented backup Path that
only avoids this faulty connection and does not
consider any other constraint condition for some
potential future failures on itself.
On the other hand, the LDC Survivability
mechanism not only avoids the fault, but it actually
configures a disjoint backup path that is both
optimal and safe. Thus, it diminishes the impact of a
further link failure (between the nodes 6 and 9), by
choosing the optical link 7-9, which is
conventionally not selected to be the minimum path
segment by the Routing Algorithm.
This has great effect on the network’s Resilience
level both for dynamic or pre planned protection
methods. In the first case, after the fault, the LDC
link state algorithm adapts to the current network
conditions and state (concerning the number of link
connections) in order to select the optimal backup
path. In the second case, again based on the same
algorithm, we pre-select this optimal path with the
advantage now of gaining in packet loss and
switchover normalization time, compared to the post
planned protection method which needs to establish
a new lightpath. The results can be a more evenly
distributed network topology and traffic as we can
clearly see in Figure 4 (Compared to Figure 1).
Figure 4: An evenly distributed network traffic
(Changwoo, 2007).
4.1 Improved Restoration Mechanism
using the LDC Algorithm
In this section we examine the issue of restoration
and calculation of a backup path after the failure has
occurred. Figure 5 illustrates the unconstrained
selection of a backup path (dashed line) in case of
failure. The particular path is in fact prone to more
potential future failures compared to the LDC
algorithm which takes into consideration the current
load for each link in the topology graph to form the
optimal path (bold line). The only difference with
the previous schemes is that this procedure is being
implemented a posterior, after the link failure event.
The results can be increased fault avoidance and
resiliency, and also a more evenly distributed
network traffic across the network, without
congested lines and nodes prone to new failures.
Figure 5: Post selecting an optimal backup path based on
the LFP factor.
5 EXPERIMENTAL RESULTS ON
NS-2
For evaluating the LDC algorithm we utilize the
network simulator ns-2 environment
(www.isi.edu/nsnam/) along with the ASONS
simulator (An Automatically Switched Optical
Network Simulator, www.telecom.ntua.gr/asons/).
For our experiments we use an example network
topology consisting of 14 nodes (figure 5). It
consists of 14 nodes and 21 STM-64 (10Gbps) SDH
Full-Duplex FiberLinks. We simulate real physical
mile distances by using equivalent link delays. Each
experiment lasts for 10 seconds, and the (sequential)
failure events occur approximately at 5.0, 5.5, 6.0,
6.5 and 7.0 sec. The failure points are between
nodes: 2-4, 4-5, 5-7, 7-8, and 8-11 respectively.
Table 1 illustrates the parameters used.
Table 1: Network parameters for the experiments.
Network Parameters
Traffic Parameters
(Exponential VBR)
Link BW : 10 Gbps
Link Delay : 5×10
-4
sec
Fiber Delay : 1×10
-3
sec
Network Load : Exponential
Traffic Rate : 100Mb
Packet Size : 100 Bytes
DCNET 2011 - International Conference on Data Communication Networking
48
Table 2: The impact of Failure Notification Distance
(relevant to the LFP Factor) and receiving Bandwidth Rate
to Recovery Time (T
REC
) and total Packet Loss (PLS).
Th
roug
h
pu
t
(Mbps)
Failure Noti
f
ication Distance
D(i,a) = 1 D(i,a) = 2 D(i,a) = 4 D(i,a) = 0
TREC PLS TREC PLS TREC PLS TREC PLS
10
4
23,5 1989 30,7 1992 42,1 3511 0,0 291
10
3
26,2 2695 33,3 4045 43,2 4112 0,0 300
10
2
26,6 2962 33,4 4858 45,6 4327 0,0 304
10 25,6 6948 36,1 9432 44,7 9878 0,0 334
In Table 2, the influence of the Failure Notification
Distance for different receiving traffic rates is being
shown. It is illustrated in this experiment that these
LSPs with higher Failure Notification Distance from
the ingress node (the node responsible for the
Failure Notification Procedure), are more likely to
experience long recovery times and packet loss.
More specifically, it is shown that T
REC
is directly
proportional to physical distance between the failure
point and the ingress node. As D(i,a), or the number
of successive hops between i (the ingress node) and
a (the hop where failure occurs), increases relatively
to the Traffic Throughput Rate, so does the
propagation link delay (T
REC
) in conjuction to the
total amount of packet loss (PLS).The final case
(D(i,a) = 0) is the most optimal since a local backup
protection method is being selected.
5.1 Evaluating the GMPLS Protection
Mechanisms using the LDC
Algorithm
For implementing the improved Protection
Mechanisms (using the LDC algorithm) we deploy
100 LSPs, in total, from nodes 1,2,3,6,9,10,12,13,14
to node 15 (egress destination node), as main
working paths. After each disjoint link failure
occurs, we compare the Default Protection state in
the asons environment, which picks up the backup
paths in an unconstrained manner, to the improved
LDC based Mechanism, which preselects the
optimal backup path(s) based on the least link-delay
constraint., We utilize the algorithm to select a pre
planned 1+1 backup path for each working path, as
well as M1 protection paths for the M:N scheme.
The two other Protection Mechanisms (1:1 and 1+N)
are equivalent to the previous ones, thus there is no
need to further examine them.
0 20406080
1
2
3
4
5
No of Sequential Link
Failures
No of Affected (Faulty) LSPs
Default
Protection
Unconstrained
Improved
Protection
LDC
Figure 6: Comparing the LDC Protection Mechanism to
the Default Unconstrained on the number of Damaged
LSPs (1+1 case).
As we can clearly see in Figure 6, for the 1+1 case,
the greater the number of sequential faults occurs the
more is the amount of LSPs being affected from the
source nodes to egress destination node. Yet, while
this true for both implementations (the
unconstrained and the metric-based), in the LDC
case we manage to obtain more lightpaths unaffected
by the network failures, thus retain traffic normality
and increased resiliency. What is most important is
that the total network traffic distribution is more
evenly applied across the topology, after the
utilization of the LDC algorithm. The results can be
increased network fault avoidance as well as reduced
packet loss, due to the unaffected and better
protected LSPs. Figure 7 also illustrates the same
effects for the M:N case.
0 10203040506070
1
2
3
4
5
No of Sequential Link Failures
No of Affected (Faulty) LSPs
Default
Protection
Unconstrained
Improved
Protection LDC
Figure 7: Comparing the LDC Protection Mechanism to
the Default Unconstrained on the number of Faulty LSP’s
(M:N case).
AN IMPROVED GMPLS SURVIVABILITY MECHANISM USING LINK DELAY-CONSTRAINED ALGORITHM
49
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
12345
Trial Number (# of Failures)
LSP Failure Ratio (%)
Improved Protection LDC Default Protection Unconstrained
Figure 8: LSP Rejection Rate Analysis for the 1+1 case
during adjacent link failures. LDC mechanism versus the
default unconstrained.
Figure 8 demonstrates the percentage of LSPs being
rejected as faulty for both Protection Mechanisms
(LDC and Unconstrained) during successive link
failures. It is important to notice that the LDC
algorithm itself increases the total ratio of functional
paths as more disjoint faults occur, contributing to
better protected and fairly distributed network
traffic. This happens because congested lines are
being avoided, the total number of traffic cut-edges
is further minimized, and finally a more even
network topology in terms of traffic flow is
achieved. Results in Figure 9 show similar
behaviour.As the Trial Number increases and more
failure events congest and aggravate the whole
network flow, the LDC solution protects more paths
than the conventional method offering higher
resilience rate and traffic smoothness.
0 500 1000 1500 2000 2500 3000
1
2
3
4
5
Improved
ProtectionLDC
Default
Protection
Unconstrained
#ofLSPs
TrialNumber(#ofFailures)
Figure 9: Number of Protected LSPs for the M:N case
during adjacent link failures. Comparing the LDC
mechanism versus the default unconstrained.
6 CONCLUSIONS FUTURE
WORK
This paper proposes the implementation of improved
GMPLS Survivability mechanisms (both Protection
and Restoration) through an efficient Constrained-
based link-state Algorithm (LDC). This algorithm
considers the delay metric on each link of the
topology, a parameter which is directly related to the
total number of connections the link has. By pre or
post selecting a path with having less connections as
the backup path, it achieves a higher safety level for
the restoration case as well as faster recovery times
and more delivered packets for the pre planned
method. For further research prospects, there will
need to be an investigation of a multi-constrained
algorithm which takes for granted other metrics and
QoS measurements, something that could provide
even more stable and robust protection across the
GMPLS networks.
REFERENCES
Calle, E.; Ripoll, J.; Segovia, J.; Vilà, P.; Manzano, M.,
2010. “A multiple failure propagation model in
GMPLS-based networks”, November-December
Changwoo Nam, Kwangsub Go, Minki Noh, Seunghae
Kim Hyuncheol Kim, Jaeyong Lee, Jinwook Chung,
2007. “LSA Expansion For Fault Recovery in GMPLS
Network”, IJCSNS International Journal of Computer
Science and Network Security, VOL.7 No.10
Dutta R., Kamal A. E., Rouskas G. N., 2008. “Traffic
Grooming for Optical Networks: Foundations,
Techniques, and Frontiers”, Springer.
Farrel A., Bryskin I., 2006. “GMPLS Architecture And
Applications”, Morgan Kaufmann Publishers.
Maier M., 2008. “Optical Switching Networks”,
Cambridge University Press.
Minei I., Lucek J., 2005. “MPLS-Enabled Applications
Emerging Developments and New Technologies”,
John Wiley And Sons Ltd.
Naoaki Yamanaka, Kohei Shiomoto, Eiji Oki, 2006.
GMPLS Technologies Broadband Backbone
Networks And Systems”, CRC Press.
Ortega E. C., 2004. “Enhanced Fault Recovery Methods
For Protected Traffic Services in GMPLS Networks
Universitat de Girona, PhD thesis.
Stern T. E., Ellinas G., Bala K., 2009. “Multiwavelength
Optical Networks”, Cambridge University Press.
DCNET 2011 - International Conference on Data Communication Networking
50