Content uploaded by Felipe Rudge Barbosa
Author content
All content in this area was uploaded by Felipe Rudge Barbosa on Aug 12, 2021
Content may be subject to copyright.
SOPO Shangai 2012 -
1
Future-proof Photonic Switched Optical Networks
for Metropolitan Access
Indayara B.Martins, Felipe Rudge Barbosa, Edson Moschim
DSIF-FEEC, University of Campinas- UNICAMP
Campinas, Brazil
e-mail: rudge@dsif.fee.unicamp.br
Abstract—Photonic Switching for traffic optimization in
Metropolitan Access Optical Networks using technologies of
optical packet/burst switching (OPS/OBS) seems to be a
future-proof solution considering high-capacity and the highly
variable traffic in these networks. In this study we propose and
analyze through computer modeling and simulations network
performance of OPS/OBS distributed switching networks and
the impact of link failure when electronic buffering at ingress
(client input) to optical network is also included. Mesh and
ring topologies are chosen as more suitable topologies and the
parameters packet loss fraction and average number of hops
are adopted for performance metrics. Minimum electronic
buffering is included at node ingress (client side) to improve
lossless access to the optical layer, yielding high throughput
and low latency. It is further observed that optical buffering is
not necessary for adequate network performance.
Keywords- Photonic Switching, Optical Packet Networks, Link-
failure, Optical Fiber Communications.
I. INTRODUCTION
A
ccess networks are client centric networks, which
today extend to the metropolitan level due to the
omnipresence of optical fibers and high-capacity systems.
Beyond that we find the highly aggregated backbone traffic.
It is clear that passive optical access networks such as Gpon
and Epon are established solutions for today´s last miles.
However, the ever-increasing amount of traffic in metro-
access networks requires ever-increasing flexibility and
availability of transport capacity [1]. On the other hand, it is
clear that WDM will remain as the best solution for
backbone traffic in the years to come [3]. Accordingly, we
believe that at the optical layer level in the metro-access
range, asynchronous OPS/OBS mesh and ring networks [2,
10,11] would cope better with packet-oriented applications;
this combined with fixed WDM and wireless integration
(3G, 4G) environments, would result as an improvement
over present-day limited bandwidth fixed star/tree
topologies [9].
This work focuses on architectures based on mesh and
ring topologies (which formally represent, respectively
closed bidimensional and closed unidimensional graphs);
the optical nodes are structured with various degrees of
interconnection. Their performances are analyzed according
to basic packet-network parameters such as average number
of hops (ANH) and packet loss fraction (PLF), applied to
optical packets. It will be seen that mesh topologies have
always better performance when compared to equivalent
rings; but ring topologies represent an enormous legacy of
SDH/SONET optical networks, and they are chosen as
means of comparison. Other topologies, such as star or tree
are not considered in this work because they are not very
capable of dealing with asynchronous traffic and self-routed
optical packets/bursts [2,10,12]. We also include effects of
link failure on network performance. Results and
methodology from previous works [4,5], are included for
comparison in the various situations with/without electronic
buffer, and with/without link failure.
II. T
OPOLOGIES AND NODE
A
RCHITECTURE
The OPS/OBS optical network architectures adopted in
the present work are based on Manhattan St. (MS) and Ring
topologies (Fig. 1). The MS mesh topology is attractive for
optical access networks because it has high connectivity and
flexibility; we adopt further distributed switching, where
optical packets are routed asynchronously through the
network, directed only by header information. We consider
networks with N nodes, with N=m
2
, where m is an integer;
for m even, we have the classical MS model; for m odd, we
have what we call quasi-regular MS (MS-q). The optical
nodes are interconnected with unidirectional links; so that
each link connects an output port from one node to the an
input port in the next node. The conventional MS is two
input and two output ports (i.e, 2x2), that is a grade-2 node;
another configuration is 3x3, that is grade-3, (Fig.1). The
4x4 configuration, grade-4, is bidirectional.
Mesh topologies have large availability of paths, and in
case of link (or node) failure a packet (or burst) can easily
change its route and follow another path; this also applies to
congestion in a given path. In the our unidirectional ring
model the nodes are connected with next neighbor node in
the clockwise direction; in counter-clockwise direction one
is jumped, and next node after neighbor (n+1) node is
connected. This avoids the next neighbor return traffic,
which happens in the classical bidirectional ring.
SOPO Shangai 2012 -
2
1
52
74
8
3 6 9
a)
b)
c]
Fig. 1: Illustrations of OPS/OBS network topologies: a] Manhattan
St. with usual grade2 and new grade3 node configuration; b] detail
of a possible grade3; c) modified ring with unidirectional links.
The detailed dynamics of the original grade-2 optical
node is described in [6], where our original experimental
demonstration was presented. The principle of operation has
been extended to the other configurations.
An optical packet arriving at the node has its header
recognition for the optical switch control electronics, a short
fiber delay line (FDL~200m, ~1µs) is included to allow for
header processing time; add/drop functionalities are also
included. An e-buffer is inserted at the client ingress of the
optical node to avoid packet loss at input; it is also necessary
to reorder the packets at egress (output) [5], because
deflections and asynchronous routing, may disorder
sequences from origin to destination. The preferred e-buffer
is limited to a short queue of ten packets to keep load/unload
times short. The optical layer itself is bufferless (except for
the short FDL) and asynchronous; that is, an optical packet is
always forwarded – either to its preferred out-port or
deflected to the available port. To resolve contention and
avoid collisions the spatial deflection routing (DR) protocol
[7] is adopted.
The optical switching for all nodes [grade-2, 3, 4] is
controlled by fast electronic logic circuits (ns rise/fall times,
[6] operating on packet-by-packet basis, determined only by
header processing (Fig.2).
Fig. 2: Optical node with grade-2 (2x2) architecture and option of
e-buffer at client ingress.
In the unidirectional links a packet from neighbor nodes i
and j follows from i to j without the possibility of leaving j
and returning back to i ; in the bidirectional, a packet may
return from j to i; causing an undesired (useless) excess of
traffic flow. On the other hand, electronic buffering (e-
buffer) at the client interface of the optical node (Fig.2) is
included to avoid PLF at input.
III.
B
ASIC
T
HEORY AND
P
ARAMETERS
If a node has k inputs and k outputs, the node
configuration is (kxk). It is assumed that the total capacity
of the networks under consideration is given by the sum of
the separate capacity of the network links, given by,
H
SNk
C
t
..
=
(1)
where
H
is the total average number of hops for the optical
packets (all packets from all origins to all destinations); N is
the number of nodes, S is the link capacity (in Gb/s); and k is
the number of input/output ports, which represents the grade
of interconnection of the optical node.
The traffic was modeled assuming uniform distribution,
that is, each node generates uniform traffic to every other
node in the network (this is also designated as the number of
traffic applications from one node to another). The effective
number of user nodes (simply “users”) in the network for
this condition is N
u
= N(N-1). We define the user-share
capacity as C
t
/N
u
, , which, using (1) becomes,
)1.(
.−
=NH
Sk
C
u
(2)
A figure of merit for network performance is defined as a
performance factor F
p
= C
t
/
H
, which can be applied to any
multihop environment that follows (1-2). It means that a
more efficient network will have higher capacity and will
have smaller average number hops (ANH); this also implies
a network with lower latency.
SOPO Shangai 2012 -
3
The calculation of ANH depends on the routing protocol
adopted [2]. If Store-and-Forward (SF) protocol is chosen
the packets/bursts always are transmitted by shortest path to
the destination address, because they follow the minimum
paths generated by Dijkstra algorithm matrix; however this
may increase latency in uncontrolled way because packets
have to wait for minimum path availability. If, on the other
hand Deflection Routing (DR) protocol is used, no buffering
at the optical layer is used and packets are immediately
forwarded to out-portt and latency is kept low and nearly
constant [2]; in this case the calculation of ANH is again
based on the minimum path matrix generated by Dijkstra
algorithm, but it is not always followed because although the
optical packet is sent by shortest path to destination,
deflections may take it through a longer path.
Table I, summarizes the analytic results of average
number of hops, network capacity and a performance factor
are presented. Transmission rate is 2.5 Gb/s, in accordance
with metro-access environment. and all the results are
obtained through the use of store-and-forward (SF) protocol,
to establish minimum values for ANH and maximum values
for C and F. Link failure is not considered, here – it will be
later. It is seen in Table I that grade3 (3x3) and grade 4
(4x4) mesh architectures are the ones with better
performance. Considering the number of nodes in each
architecture, the ANH increases gradually, with the 16-node
having the highest relative value; the capacity
C
t
increases
approximately linearly, but again on a linear plot the 16-
node performs below the others. Therefore, on a network
planning perspective two 9-node networks interconnected
by a higher order node (a local area head-end), would
perform much better than a 16-node network.
Another highly significant parameter is the packet loss
fraction (PLF), defined as the ratio,
prp
PLF
−
=
(3)
where p is the number of packets generated in a given node
and r is the number of packets actually received.
To further analyze impact of link failure, equation (1) for
the capacity is modified accordingly:
H
SfNk
C
t
)..(
−
=
(4)
with f the number of link failures involved. The cases of uni-
and bi-directional links are separately considered.
IV. I
MPACT OF
L
INK
F
AILURE
In high capacity optical networks it is important to have
network protection mechanisms to prevent failures from
disrupting network services. We consider impact of single
link failures, due cuts of fibers and cables, or in
transmitter/receiver pairs, on the parameters of the Table I.
Table I – Evaluation of Network parameters for mesh and ring
topologies with unidirectional and bidirectional links.
Average Number
of hops
H
Perform.
Factor F
p
Topology
N
Netwk.
Capac.
C
t
(Gb/s)
SF DR SF DR
MSq-9 9 22,3
2,0
2,2
11,1
10
MS-16 16 27,3
2,9
3,7
9,3
7,2
MSq-25 25 38,1
3,3
3,9
11,6
9,7
Grau 2
MS-36 36 48,5
3,7
4,5
13
10,7
MSq-9 9 40,8
1,6
1,3
24,7
32,3
MS-16 16 56,0
2,1
1,6
26,2
34,3
MSq-25 25 73,2
2,5
2,0
28,6
35,3
Grau 3
MS-36 36 92,1
2,9
2,4
31,4
37,7
MSq-9 9 60
1,5
1,5
40
39,7
MS-16 16 75
2,1
2,2
35,1
33,3
MSq-25 25 100
2,5
2,6
40
38,3
Grau 4
MS-36 36 116
3,1
3,3
37,8
35,3
Ring-9(bid) 9 18
2,5
2,7
7,2
6,6
Ring-9 (uni) 9 20
2,2
2,5
9,1
7,9
Ring-16(bid) 16 18,7
4,2
4,6
4,4
4,0
Ring-16(uni) 16 23,0
3,5
3,7
6,6
6,2
Ring-25(bid) 25 19,2
6,5
6,9
2,9
2,7
Grau 2
Ring25 (uni) 25 25,8
4.8
5,3
5,3
4,8
Fig. 3 illustrates the readjustment of traffic distribution for
mesh in 2x2 node configuration before and after a major
link failure. Applications represent link usage. We choose
more used links in order to set an upper bound for the
failure impact. It is observed that in the failure mode traffic
is redistributed, and the average number of applications per
link increases in the neighborhood of the failure, as
expected; an evaluation of the time evolution of the
distribuition is planned. We have also evaluated impact on
capacity, which lowers less than 10% for mesh architetcures
and more than 20% for the rings.The details of protection
mechanisms is beyond the scope of this work.
a]
(14,13)
SOPO Shangai 2012 -
4
b]
Fig.3 - Link utilization MS-16 (2x2): (a) normal operation; and (b)
with link failure.
Similarly, Fig. 4 shows results for ring (2x2). Again we
notice that the unidirectional ring is not severely affected by
a single failure. A significant difference would be the case
of a double failure, in which a ring may collapse depending
on the relative locations, whereas a mesh would do not.
Double or higher link failures need not be considered here
because their probability of occurrence is various orders of
magnitude smaller than single failure [2,Pnet].
a]
b]
Fig.4 - Link utilization Ring-16 unid: (a) normal operation; and (b)
with link failure
V. T
RAFFIC
S
IMULATION
M
ETHODS AND
N
ETWORK
C
ONFIGURATIONS
This section discusses the procedures and conditions for
simulations.
Fig. 5: Simulation Dynamics for SF and DR.
Fig. 5 presents a diagram of the simulation dynamics for
both SF and DR protocols.
The network model uses unidirectional links, with uniform
traffic distribution (every node generates the same amount of
traffic to every other node) during each simulation round
time (20ms); the intervals between optical packets vary from
0.1 to 1 packet duration. Every connection is set up with
UDP protocol, so that lost packets are not retransmitted. The
optical packets have fixed-size of 500 bytes; note that they
are easily extensible to 10 or 20 times larger for burst
switching, but then the network performance has to be
reevaluated. Transmission rate is 2.5 Gb/s, in accordance
with metro-access environment. The total number of packets
generated for simulation rounds is 2x10
5
. Bit-error rate
(BER) is assumed 10
-9
. Link length is 10km for all links.
Note that the ingress e-buffer implementation appears only
for the DR branch; the SF protocol always uses unlimited
buffering, which increases network latency as previously
discussed.
Analytic results and data processing and plotting, were
carried out with an open version Matlab®; whereas for
traffic simulations the Network Simulator (NS-2) was used.
VI. N
ETWORK
R
ESULTS AND
D
ISCUSSION
In this section compare results of packet loss fraction
(PLF) and the average number of hops (ANH) for MS-16
and Ring-16 node networks. As discussed before, the other
networks tend to perform better. We show the impact of link
failure on these parameters when a failure occurs in the more
used links, and compare these parameters wnhe e-buffer at
node ingress is implemented.
In Fig. 6a we observe that implementing e-buffers has a
dramatic impact on the packet loss fraction (PLF) which
drops markedly and remains low even for high loads. The
crossing behaviour is attributed to the e-buffer load/unload
times, which vary as traffic load varies [5, Bonani] from
lower to higher load. In Fig.6b one can see that the ANH is
higher with the presence of e-buffers, because the average
number of packets circulating in the network is higher as the
PLF is much lower (Fig.6a).
(14,15)
SOPO Shangai 2012 -
5
a]
b]
Fig. 6 : MS-16 (2x2) :: a) PLF ; b) ANH; both with and without e-
buffer, and with/without failure.
Fig.7 : PLF for Ring-16 2x2 unidirectional with/without e-buffer
and link failure.
When ingress buffering is not used, almost all the PLF
occurs at the optical layer input (add-port) due to collisions.
It is interesting that the NS has an internal decision of
preserving the line packets over the incoming ones.
Results for mesh indicate that networks with grade-3
nodes are those that tend to have smallest ANH because
they show the largest number of alternative paths between
different node pairs. PLF is not affected by this feature.
Finally, in Fig. 7 the PLF for the modified Ring (grade-2),
with/ and without e-buffers and occurrence of link failure.
We notice that with e-buffers PLF becomes very low or
negligible. The crossing at higher loads is attributed to
load/unload times of buffers, which change the traffic load.
VII. C
ONCLUSION
We have evaluated and compared mesh and ring optical
network architectures, through quantitative analytic and
simulation results using models with unidirectional and
bidirectional links. This work extends, modifies and
innovates on previous results. We implemented e-buffers at
client node ingress and performed analyses based on
paratmeters of network capacity, average number of hops
and packet loss fraction for various network configurations.
Impact of single link failures for all topologies was also
considered and demonstrated that these network topologies
(mesh and ring) are not strongly affected by link failures. As
expected, the quantitative evaluation showed improved
performance in all cases, which confirms the present
methodology as useful tool for network planning and
design. Future work will consider topologies other than
regular and quasi-regular ones studied here. We also plan to
include cost of link usage and occurrence of failure of a
node. A comparison with other solutions for OBS networks
[12] is also planned.
R
EFERENCES
[1] C. Develder, A. Stavdas, Fabio Neri, J. Solé-Pareta, N. Le
Sauze, and P. Demeester, “Benchmarking and Viability
Assessment of Optical Packet Switching for Metro
Networks,” J. Lightwave Tech., Vol. 22, no. 11, Nov. 2004.
[2] I. B. Martins, F. Rudge Barbosa ; L. H. Bonani ; E. Moschim .
“Evaluation of Throughput and Protection in Optically
Switched Metropolitan Networks Architectures”, Photonic
Network Communications PNET, number 2, April 2010. DOI:
10.1007/s11107-010-0249-z “PNET 2010 .
[3] R. Ramaswami, K. N. Sivarajan, “Optical Networks: a
practical perspective”, Morgan Kaufmann, Academic Press,
Boston, USA 3
rd
. Ed., 2010.
[4] I. B. Martins, F. Rudge Barbosa, L.H.Bonani, E. Moschim,
“A Study in Photonic Switched Networks considering Link
Failure and Ingress Buffering”. Proc. 34
th
IEEE Sarnoff
Symposium, May 2011, Princeton- EUA
[5] L.H. Bonani, F. Rudge Barbosa, E. Moschim, and R. Arthur,
“Analysis of Electronic Buffers in Optical Packet/Burst
Switched Mesh Networks,” Intl. Conference on Transparent
Optical Networks-ICTON-2008, June 2008 – Athens, Greece.
[6] F. Rudge Barbosa, D. Maia, and E. Moschim, “Optical Packet
Switching Node for Metro-Access Networks,” paper PD-160,
Proceed. 29th. ECOC’2003, Sept. 2003, Rimini, Italy.
[7] S. Yao, B. Mukherjee, S. J. Yoo, and S. Dixit, “Unified Study
of Contention Resolution in Optical Packet Switching
SOPO Shangai 2012 -
6
Networks,” IEEE J. Lightwave Tech., vol.21, no.3, p.672,
March 2003.
[8] D. A. Schupke and R. Prinz. “Capacity Efficiency and
Restorability of Path Protection and Rerouting in WDM
Networks Subject to Dual Failures”, Photonic Network Comm
PNET, Vol 8, number 2, p.191, Springer, Netherlands Sept.
2004
[9] Rec. ITU-T 984.1, Gigabit-capable passive optical networks
(GPON): General characteristics, March 2008.
[10] I.Tomkos, S. Azodolmolky, D. Klonidis, M. Aggelou, and K,
Margariti “Dynamic Impairment Aware Networking for
Transparent Mesh Optical Networks: the EU Project
DICONET,” Proc. ICTON’2008, June 2008- Athens, Greece.
[11] M. Maier , Optical Switching Networks (Cambridge
University Press, 2008);
[12] E. Mobilon, et al; “Hardware Architecture for Optical Packet
and Burst Switching Applications”, Proceeds of 8
th
Intl.
Telecomm Symp.: ITS 2006; Jhongli, Taiwan, p278, June
2006 . ISBN 85-89748-049
~~ ~~ ~~