Está en la página 1de 9

Photonic Network Communications, 6:1, 3341, 2003

# 2003 Kluwer Academic Publishers. Manufactured in The Netherlands.

Connection Establishment of Label Switched Paths in IP/MPLS over


Optical Networks
Xiaomei Niu, Wen-De Zhong*
School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore
E-mail: ewdzhong@ntu.edu.sg

Gangxiang Shen
University of Alberta, Canada
E-mail: gshen@edm.trlabs.ca

Tee Hiang Cheng


Institute for Infocomm Research, Singapore
E-mail: ethcheng@ntu.edu.sg
Received November 13, 2002; Revised January 30, 2003; Accepted February 3, 2003

Abstract. We propose and investigate three connection admission control policies for the establishment of label switched paths (LSPs) in IP/
MPLS over optical networks. We show that the policy of establishing LSPs rst in the optical layer achieves a better blocking performance. We
examine the effect of the number of add/drop ports of optical cross-connects (OXCs) on the LSP blocking performance. We show that there
exists a lower bound for the number of add/drop ports of OXCs for the network to achieve almost the best LSP blocking performance.
Keywords: IP/MPLS over optical networks, label switching router, label switched path, lightpath, connection admission control

Introduction

It is anticipated that the merging of the Internet and


the optical network will be the key to build the next
generation telecommunications network infrastructure [16]. However, the conventional Internet
network was not designed to provide guaranteed
quality of service (QoS) to the end users. Multiprotocol label switching (MPLS) [23] and generalized MPLS (GMPLS) [4] have been proposed to
provide circuit-switching-like guaranteed services to
the end user. Within the framework of IP/MPLS over
optical networks, a label switched path is established
between two label switching routers (LSRs) before
packets are transmitted between them. Recently, most
studies on IP/MPLS over optical networks have been
focused on network control and management protocols [1], and the routing and wavelength assignment
(RWA) problems of optical transport networks.
*Corresponding author.

Studies of the RWA problem have been carried out


under both static and dynamical trafc demands with
the aim of minimizing the required number of
wavelengths and optimizing the optical network
performance [712]. However, little work has been
done on the network performance evaluation when
both the IP/MPLS layer and the optical layer are
considered together in a holistic manner.
In this paper, based on an overlay model, we
propose and examine three connection admission
control (CAC) policies for the establishment of label
switched paths (LSPs). Aiming to reduce the number
of add/drop ports of optical cross-connects (OXCs),
we investigate the effect of the number of add/drop
ports of OXCs on the LSP blocking performance. We
show that there exists a lower bound for the number of
add/drop ports of OXCs for the network to achieve
almost the best LSP blocking performance. In Section
2, we describe the network and trafc models used in

34

X. Niu et al./Connection Establishment of Label Switched Paths

this study. The CAC policies are presented in Section


3, and routing of lightpaths and LSPs in the overlay
model is described in Section 4. In Section 5, we
examine the LSP blocking performance for different
CAC policies. The effect of the number of add/drop
ports of OXCs on the LSP blocking performance is
presented in Section 6. Finally we give our conclusion
in Section 7.
2

Network and Trafc Models

A typical architecture of IP/MPLS over optical


networks consists of two layers. One is the IP/MPLS
layer, which consists of IP/MPLS label switching
routers (LSRs). The other is the optical layer, which is
comprised of OXC nodes interconnected by ber
links. Based on the relationship of the control planes
in the two layers, the architecture of IP/MPLS over
optical networks is generally classied into three
interconnection models, namely the overlay model,
the augmented model and the peer model [2]. In this
paper, we focus on the overlay model, which is the
most practical model in the near future. Under the
overlay model, the control planes of the IP/MPLS
layer and the optical layer have the loosest relationship. The routing instances, the topology information
distribution and the signaling processes in the two
layers are independent of each other. The communication between the two layers is very limited, except
that service requests and responses (e.g., lightpath

establishment, release, and modication) are


exchanged through the user-network-interface
(UNI). Fig. 1 illustrates the overlay model used in
our study. In this model, a point-to-point path between
two LSRs in the IP/MPLS layer is called a virtual IP/
MPLS path, which is created over an established
lightpath in the optical layer. As shown in Fig. 1, a
network node consists of an OXC and an LSR with a
UNI between them for information exchange. There
are add/drop ports on the OXC. Basically, the drop
ports are necessary to receive trafc terminating at the
local LSR and the add ports are used to transmit trafc
originating at the local LSR. In addition to this basic
function, the add/drop ports can also be used to deliver
transit trafc that does not terminate at the local LSR
so as to enhance the LSP blocking performance. The
effect of the number of add/drop ports on the LSP
blocking performance will be examined in Section 6.
We assume a dynamic trafc model for our study.
The trafc load is in the form of LSP requests, which
arrive at the IP/MPLS layer. The required bandwidth
of each LSP request is a random variable, which is
uniformly distributed between 0 and C, where C is the
capacity of a single wavelength. LSP request arrivals
follow a Poisson distribution, and each of them
requires a duplex connection between a pair of LSR
nodes. The holding time of each LSP connection has a
negative exponential distribution. The trafc load is
uniformly distributed in the network. That is, an LSP
request originating at a node is equally likely to be
addressed to any of the other nodes in the network.

Fig. 1. Network model and the node architecture for IP/MPLS over optical networks.

X. Niu et al./Connection Establishment of Label Switched Paths

Based on the above assumption, we dene the


normalized trafc load generated at each LSR node,
r, as follows:
r

l6m6average bandwidth of LSP requests


;
C

where l is the LSP arrival rate at an LSR node, and m


is the average holding time of LSP requests. The
normalized trafc load is in the unit of wavelength.
The connections of LSP requests are established and
released dynamically. We use request blocking ratio
(RBR) as the performance evaluation parameter. The
RBR is dened as the ratio of the accumulated
bandwidth of the blocked LSP requests to the
accumulated bandwidth of the total LSP requests,
which is expressed as:
P
BBRi
RBR Pi
;
1
j BRj
where BBRi is the bandwidth of the i-th blocked LSP
request and BRj is the bandwidth of the j-th LSP
request.

Connection Admission Control Policies

In the overlay model, either the IP/MPLS layer or the


optical layer can be selected for the establishment of
LSPs. According to the required bandwidth of LSPs
and the available network resource, there are a few
possible ways to set up LSPs. For example, in Fig. 1, a
new LSP request between LSR1 and LSR3 may be
established on a one-hop route of the existing virtual
IP/MPLS path of LSR1-LSR3, or a multi-hop route of
LSR1-LSR4-LSR3 consisting of two existing virtual
IP/MPLS paths in the IP/MPLS layer. Alternatively, a
new lightpath of OXC1-OXC2-OXC3 in the optical
layer can be initiated to create a new virtual IP/MPLS
path of LSR1-LSR3 in the IP/MPLS layer, and the
new LSP request will be established on this new
virtual IP/MPLS path. Which layer will be the rst
choice to establish the LSPs is the issue of CAC
policies. We here consider three CAC policies and
present their performance results in Section 5.
3.1 IP/MPLS-Layer-First Policy
With this policy, the IP/MPLS layer is always
attempted rst when a new LSP request arrives. If
there is sufcient remaining capacity in the IP/MPLS

35

layer, the new LSP request will be set up in it. If this


attempt fails, the LSP request is transferred through
the UNI to the optical layer for a second try. If there
are sufcient free wavelengths in the optical layer, a
new lightpath is set up. Meanwhile, a new virtual IP/
MPLS path is created in the IP/MPLS layer and the
new LSP is established on it. Once the new LSP is set
up successfully, the network resource state should be
updated subsequently. However, if both attempts fail,
then the new LSP request is blocked.

3.2 Optical-Layer-First Policy


Under this policy, the optical layer always attempts to
set up a new lightpath rst when a new LSP request
arrives. If this attempt succeeds, a new virtual IP/
MPLS path is created in the IP/MPLS layer and the
new LSP is set up on it. If the rst attempt fails, this
LSP is forwarded to the IP/MPLS layer for a second
attempt. If there is enough remaining bandwidth on
the existing virtual IP/MPLS paths to support it, the
new LSP will be set up; otherwise it will be blocked.
Similarly, each time the new LSP is established
successfully, the network resource state should be
changed accordingly.

3.3 Lightpath-Establishing-Threshold-Based
Policy
In the above two CAC policies, we try to set up
connections for all LSPs either in the IP/MPLS layer
rst or in the optical layer rst. As a compromise of
the rst two CAC policies, we propose a lightpathestablishing-threshold (LET)-based policy. We introduce a parameter, LET, which will determine which
layer should be attempted rst to support the LSP
requests. When a new LSP request arrives and the
ratio of its bandwidth request to the capacity of a
single wavelength is higher than the LET value, the
optical layer is selected as the rst choice; otherwise,
the IP/MPLS layer is selected. When the LET value is
set to 0, the LET-based policy becomes the opticallayer-rst policy; and when LET value is set to 1, it
becomes the IP/MPLS-layer-rst policy. Our purpose
is to study the effect of the LET values on the network
performance and to see if it is necessary to classify the
LSP requests according to their required bandwidths
in order to achieve the best blocking performance.

36
4

X. Niu et al./Connection Establishment of Label Switched Paths

Routing of Lightpath and LSP

The lightpath routing in the optical layer is based on


Dijkstra's shortest path algorithm [13] and the xed
topology of the optical layer. It is noted that the xed
topology of the optical layer is composed of OXC
nodes and ber links. We assume the weight of each
ber link is identical, regardless of the distance
difference among all the ber links. Let N be the
number of OXC nodes in the optical layer, and let
A [aij] be an N 6 N matrix representing the xed
topology of the optical layer, where

aij aji

8
1;
>
<
>
:

?;

if there is a fiber link


between OXC node i and
OXC node j, and i = j
otherwise.

Based on the xed topology of the optical layer A, a


lightpath request between two OXC nodes will be set
up on the route with the minimum number of hops.
This routing algorithm is the simplest to implement,
since the routes of all lightpaths between any two
OXC nodes are xed.
The LSP routing in the IP/MPLS layer is also based
on Dijkstra's shortest path algorithm and the dynamic
topology of the IP/MPLS layer, which is composed of
LSRs and virtual IP/MPLS paths. The topology of
IP/MPLS layer is not xed, but changing dynamically.
The reasons are given below. Firstly, all virtual
IP/MPLS paths are set up and released together with
their corresponding lightpaths. Secondly, the virtual
IP/MPLS paths selected to support each LSP request
could be different. In our study, we assume that there
are multiple wavelengths on each ber link. Thus it is
possible that more than one lightpath, i.e., more than

one IP/MPLS virtual path, may co-exist between a


pair of LSR nodes. For each LSR node pair where
there are more than one established virtual IP/MPLS
paths, we will check upon a new LSP request arrival
the virtual IP/MPLS paths in order, and select the rst
virtual IP/MPLS path whose spare capacity can
support the bandwidth request of the new LSP as the
path to form a new dynamic topology. We assume the
weight of all virtual IP/MPLS paths is identical. Let N
be the number of the LSR nodes in the IP/MPLS layer
and B bij represents an N 6 N matrix of the
dynamic topology of the IP/MPLS layer, where
8 1; if there exists a virtual
>
>
>
IP/MPLS path between LSR
>
>
>
>
node i and LSR node j,
<
whose remaining capacity can
bij bji
>
>
satisfy the bandwidth request
>
>
>
>
>
of this LSP, and i 6 j
:
?; otherwise
Based on this dynamic topology of the IP/MPLS layer
B, a route with a minimum number of hops can be
computed for the new LSP connection.

LSP Blocking Performance

The LSP blocking performance is evaluated by


computer simulation on two typical network topologies as shown in Fig. 2: the ARPA2 network and the
NSFNET network [14]. The ARPA2 network consists
of 21 nodes and 25 ber links; while the NEFNET
network has 14 nodes and 21 ber links. Unless
explicitly indicated, in the simulation study we
assume that each ber link has 16 wavelengths in

Fig. 2. Two typical network topologies.

X. Niu et al./Connection Establishment of Label Switched Paths

the ARPA2 network and eight wavelengths in the


NSFNET network. In this section, we do not place
restriction on the number of add-drop ports on the
OXCs so as to obtain the best blocking performance.
The effect of the number of add/drop ports on the LSP
blocking performance will be examined in the next
section.
According to whether OXCs are capable of
wavelength conversion or not, we classify the networks into two categories in our study: the networks
with no wavelength conversion capability (NWC),
where all OXC nodes have no wavelength conversion
capability; and the network with full wavelength
conversion capability (FWC), where each OXC node
is able to convert any wavelength on an input port to
any other wavelength on an output port. When a new
lightpath request arrives in an FWC network, the
network only needs to judge if there are free
wavelengths along the xed route of this lightpath.
When a new lightpath request arrives in an NWC
network, the restriction of wavelength continuity
should be considered for the lightpath establishment
in the optical layer, namely the lightpath must use the
same wavelength all through its route from source to
destination. In our study, for each lightpath request in
the NWC networks, we check the wavelengths on its
xed route in order and select the rst free wavelength
to set up this lightpath.
Figs 3 and 4 show the RBR versus the normalized
trafc load per node for the rst two CAC policies in
both the ARPA2 and NSFNET networks. It can be
seen that the optical-layer-rst policy achieves a
lower blocking probability than the IP/MPLS-layerrst policy for both networks with FWC and NWC.
This can be explained as follows. When an LSP is set
up rst in the optical layer, the connection to be
established is a one-hop, source-to-destination virtual
IP/MPLS path in the IP/MPLS layer, which occupies
only one lightpath in the optical layer. While it is
more likely to be a multi-hop route in the IP/MPLS
layer, which will occupy more than one lightpath in
the optical layer and hence consume more network
resource, if an LSP is established rst in the IP/MPLS
layer. Since the total network bandwidth resource is
nite when the network topology and the number of
wavelengths on each ber are xed, if LSPs are
always established rst in a resource-consumptive
way, the network resource utilization is lower and
hence the blocking probability is higher. Thus the
optical-layer-rst policy has better performance.

37

Fig. 3. Request blocking ratio versus normalized trafc load per


node in the NSFNET network with FWC and NWC.

Figs 5 and 6 show the RBR versus the normalized


trafc load per node for the LET-based policy in both
the ARPA2 and NSFNET networks. The LET-based
policy is intended to achieve a better performance.
However, to our surprise, the change of the LET
values has little inuence on the LSP blocking
performance. As shown in the two gures, the
curves of LSP blocking probability versus the LET
value remain approximately at or increase slightly as
the LET value increases. There are not obvious
minimum blocking probability points in the middle of
these curves. This means that classifying the LSP
requests according to their required bandwidths has
little impact on the LSP blocking performance
improvement. Thus, there is no need to classify LSP
requests, and all LSP requests can be treated equally.

Fig. 4. Request blocking ratio versus normalized trafc load per


node in the ARPA2 network with FWC and NWC.

38

X. Niu et al./Connection Establishment of Label Switched Paths

Fig. 5. Request blocking ratio versus LET value in the NSFNET


network with FWC and NWC for different values of normalized
trafc load per node.

Fig. 7. Request blocking ratio versus add/drop ratio for the


NSFNET network with FWC for different values of normalized
trafc load per node.

performance, we introduce a parameter called add/


drop ratio, which is dened as the ratio of the number
of add/drop ports per ber to the number of
wavelengths per ber. Our aim is to nd the minimum
required number of add-drop ports on OXCs and
meanwhile to achieve the best blocking performance.
Figs 710 show the relationship between the RBR
and the add/drop ratio for both the NSFNET and
ARPA2 networks with FWC and NWC respectively.
As shown in these gures, for both networks with
FWC and NWC, when the add/drop ratio is small, the
RBR drops rapidly as the add/drop ratio increases.
When the add/drop ratio reaches a certain value and
further increases, the RBR remains almost unchanged.
This means that there is a lower bound for the

Fig. 6. Request blocking ratio versus LET value in the ARPA2


network with FWC and NWC for different values of normalized
trafc load per node.

6 Effect of the Number of Add/Drop Ports on


LSP Blocking Performance
We now examine the impact of the number of add/
drop ports on the LSP blocking performance. A
lightpath (and hence a virtual IP/MPLS path) may not
be able to be established between two LSR nodes,
even if there are enough free wavelengths between
their associated OXC nodes when the number of add/
drop ports on OXC nodes is limited. On the other
hand, it could be too costly if all the wavelengths are
added/dropped at all OXC nodes. To study the effect
of the number of add-drop ports on the LSP blocking

Fig. 8. Request blocking ratio versus add/drop ratio for the


NSFNET network with NWC for different values of normalized
trafc load per node.

X. Niu et al./Connection Establishment of Label Switched Paths

Fig. 9. Request blocking ratio versus add/drop ratio for the ARPA2
network with FWC for different values of normalized trafc load
per node.

39

Fig. 11. Request blocking ratio versus add/drop ratio for various
numbers of wavelengths per ber for the NSFNET network with
FWC (optical-layer-rst policy adopted, and r 0.975).

add/drop ratio of OXCs, where the network can


achieve a blocking performance as good as that with
100% add/drop capability. This is because a portion of
trafc just bypasses OXC nodes and does not need to
be added/dropped. It is found that the lower bound for
the add/drop ratio is almost not affected by the factors
of trafc load and wavelength conversion capability.
The lower bound for the add/drop ratio could be used
in determining the minimum required number of add/
drop ports on OXCs in order to achieve almost the
best blocking performance without paying unnecessary network cost.
We now change the number of wavelengths per
ber to investigate the effect of the add/drop ratio on

the LSP blocking performance. Here we select the


optical-layer-rst policy. Figs 1114 show the RBR
versus the add/drop ratio for different numbers of
wavelengths per ber for both NSFNET and ARPA2
networks. In Figs 1114, the parameter W is the
number of wavelengths per ber. From these gures,
we can see that when the add/drop ratio is relatively
high, the more the number of wavelengths per ber,
and the better the network blocking performance is.
But it is not true when the add/drop ratio is low. In
addition, as shown in the four gures, the number of
wavelengths per ber affects slightly the lower bound
for the add/drop ratio for both the NSFNET and
ARPA2 networks with FWC and NWC.

Fig. 10. Request blocking ratio versus add/drop ratio for the
ARPA2 network with NWC for different values of normalized
trafc load per node.

Fig. 12. Request blocking ratio versus add/drop ratio for various
numbers of wavelengths per ber for the NSFNET network with
NWC (optical-layer-rst policy adopted, and r 0.975).

40

X. Niu et al./Connection Establishment of Label Switched Paths

best LSP blocking performance, we have investigated


the effect of the number of add/drop ports of OXCs on
the blocking performance. It has been found that there
is a lower bound for the add/drop ratio for the
networks to achieve almost the best blocking
performance with fewer add/drop ports. This lower
bound value should be used in determining the
minimum required number of add/drop ports on the
OXC nodes so that the network can achieve almost the
best blocking performance but without unnecessarily
increasing the network cost.

Fig. 13. Request blocking ratio versus add/drop ratio for various
numbers of wavelengths per ber for the ARPA2 network with
FWC (optical-layer-rst policy adopted, and r 1).

Fig. 14. Request blocking ratio versus add/drop ratio for various
numbers of wavelengths per ber for the ARPA2 network with
NWC (optical-layer-rst policy adopted, and r 1).

Conclusion

We have proposed and investigated three connection


admission control policies for IP/MPLS over optical
networks under the overlay model. Simulation results
have revealed that the optical-layer-rst policy
achieves a better blocking performance than the IP/
MPLS-layer-rst policy. It has been shown that the
change of LET values does not lead to much
improvement in the LSP blocking performance, and
hence there is no need to classify the LSP requests
according to their required bandwidths. Aiming to
reduce the network cost and meanwhile to achieve the

References
[1] G. M. Bernstein, J. Yates, D. Saha, IP/MPLS-centric control
and management of optical transport networks, IEEE
Commun. Mag., vol. 38, no. 10, (Oct. 2000), pp. 161167.
[2] B. Rajagopalan et al., IP/MPLS over optical networks: A
framework, IETF Internet draft, (Sept. 2001).
[3] B. Davie, Y. Rekhter, MPLS technology and applications,
(Morgan Kaufmann, 2000).
[4] O. Aboul-Magd et al., A framework for generalized multiprotocol label switching, IETF Internet draft, (Sept. 2001).
[5] Y. Ye, S. Dixit, M. Ali, On joint protection/restoration in IP/
MPLS-centric DWDM-based optical transport networks,
IEEE Commun. Mag., vol. 38, no. 6, (June 2000), pp. 174
183.
[6] D. O. Awduche, MPLS and trafc engineering in IP/MPLS
networks, IEEE Commun. Mag., vol. 37, no. 12, (Dec. 1999),
pp. 4247.
[7] R. Ramaswami, K. N. Sivarajan, Routing and wavelength
assignment in all optical networks, IEEE/ACM, Trans.
Networking, vol. 3, no. 5, (Oct. 1995), pp. 489500.
[8] G. Shen, S. K. Bose, T. H. Cheng, C. Lu, T. Y. Chai, Efcient
heuristic algorithms for light-path routing and wavelength
assignment in WDM networks under dynamically varying
loads, Computer Communications, vol. 24, (2001), pp. 364
373.
[9] I. Chlamtac, A. Farago, T. Zhang, Lightpath (wavelength)
routing in large WDM networks, IEEE J. Select. Areas
Commun., vol. 14, no. 5, (June 1996), pp. 909913.
[10] D. Banerjee, B. Mukherjee, A practical approach for routing
and wavelength assignment in large wavelength-routed optical
networks, IEEE J. Select. Areas Commun., vol. 14, no. 5, (June
1996), pp. 903908.
[11] K. C. Lee, V. O. K. Li, A wavelength routing algorithm in
wide-area all-optical networks, IEEE J. Lightwave Technol.,
vol. 14, no. 6, (June 1996), pp. 12181229.
[12] R. Ramamurthy et al., Capacity performance of dynamic
provisioning in optical networks, IEEE J. Lightwave Technol.,
vol. 19, no. 1, (Jan. 2001), pp. 4048.
[13] Ronald Gould: Graph theory, the Benjamin/Cummings
Publishing Company, Inc., (1988).
[14] O. Crochat, J. Boudec, Design protection for WDM optical
networks, IEEE J. Select. Areas Commun., vol. 16, no. 7,
(Sept. 1998), pp. 11581165.

X. Niu et al./Connection Establishment of Label Switched Paths


Xiaomei Niu received her B.E. degree
from Beijing University of Posts and
Telecommunications, Beijing, China, in
1995. From 19952000, she worked at
the Network Department, Science and
Technology Information Institute, Beijing,
China. She was pursuing her postgraduate
study towards a Master degree at Nanyang
Technological University, Singapore, from
2000 to 2002, and has submitted her thesis
recently. She is currently an academic
visitor with the University of Western Australia, Australia, where
she is working in the area of communication networks. Her research
interests include telecommunication network analysis and optical
networking.
Wen-De Zhong is an associate professor
with School of Electrical and Electronic
Engineering, Nanyang Technological
University (NTU), Singapore. He received
his Ph.D. degree from the University of
Electro-Communications, Japan, in 1993.
Since then, he has been actively involved in
research and development of optical communication systems and networks. From
19931995, he was a postdoctoral fellow
at NTT Network Service Systems
Laboratories, NTT, Japan, where he was working on design and
development of photonic switching systems. From 19952000, he
was a senior research fellow with the Australian Photonics
Cooperative Research Centre, Photonics Research Laboratory at
the University of Melbourne, Australia, where he was working on
design and analysis of WDM systems and networks. He has
published more than 80 technical papers in prestigious journals and
international conferences. His current research interest includes
WDM systems and networks, IP over optical networks, and
photonic packet switching systems.

41

Tee Hiang Cheng has been on the


academic staff of Nanyang Technological
University (NTU) for the past 10 years. He
was previously Director of the Network
Technology Research Centre in NTU.
Presently he is on a two-year secondment
to the Institute of Infocomm Research to
lead its Optical Communication and
Networking Research. He has been actively involved in broadband
and optical networking research. He has published more than a
hundred technical papers in prestigious international journals and
conferences in these two areas. He also has research interest in
wireless local area network technology and co-founded RFNet
Technology Pte Ltd, which specializes in designing broadband
wireless networking products. He is very active in providing
consultancy and professional services to the industry. Presently, he
is Chairman of the IEEE Communications Chapter for Singapore,
Technical Activities Coordinator for IEEE Region 10, and Vice
President of Broadband SIG (Singapore)

Gangxiang Shen obtained his B.Eng.


degree in 1997 from Zhejiang University,
P.R. China, and M.Sc. degree from
Nanyang
Technological
University,
Singapore, in 1999. After that, he joined
the laboratory of the Network Technology
Research Centre (NTRC) of Nanyang
Technological University as a Research
Associate, then the Institute for
Communication Research (ICR) of
Singapore as a Senior Research Engineer.
He is currently with Network System Group of TRLabs and
University of Alberta, Canada, persuing a Ph.D. His main research
interests focus on all-optical networks and survivable networks. He
is a student member of IEEE.

También podría gustarte