STATISTICAL TRAFFIC MULTIPLEXING WITH SERVICE

GUARANTEES OVER OPTICAL CORE NETWORKS

A. Drakos, T. Orphanoudakis, C. (T) Politi and A. Stavdas

Department of Telecommunications Science and Technology, University of Peloponnese, Karaiskaki Street

Tripolis 22100, Greece

Keywords: Optical Networks, Optical Communications, Statistical Multiplexing, OBS, CANON.

Abstract: Statistical multiplexing at the optical layer has been considered a critical requirement in order to build the

next generation of ultra-high capacity optical transport networks in a cost-efficient manner. However, even

today, the state of the art of commercially available optical core networks is based on mature wavelength

switching and routing technologies, which lack a transport and control plane architecture that can support

statistical traffic multiplexing with guaranteed Quality of Service (QoS) across a wide range of QoS

parameters even if they can support fast reconfiguration at msec time scales. For several years, most

research efforts have focused on the concepts of Optical Burst Switching (OBS) and Optical Packet

Switching (OPS), which are based on the hybrid use of electronic nodes and optical switches to exploit

Time Division Multiplexing (TDM) in order to achieve statistical multiplexing and dynamic resource

reservation over optical networks. While burst switching has been experimentally proven as a technically

feasible technique, its performance suffers especially under strict requirements for QoS guarantees. In this

paper we evaluate the performance gains that can be achieved exploiting statistical multiplexing over a large

scale core optical network and we demonstrate the efficiency of the CANON architecture (Clustered

Architecture for Nodes in an Optical Network) as a viable alternative to OBS, which can achieve both

targets for statistical multiplexing gains and QoS guarantees at the same time.

1 INTRODUCTION

Latest trends in backbone network traffic

demonstrate that apart form the tremendous increase

in volume, modern telecommunication services

result in traffic patterns with increased dynamics in

the form of spatial and temporal asymmetry and

stringent requirements for Quality of Service (QoS)

guarantees. While the traffic growth can be

accommodated by exploiting the capacity of WDM

systems, the available fiber and wavelength

resources will not be able to serve the constantly

increasing demand unless the reservation of these

resources is performed in an efficient manner.

To address these requirements one approach is to

deploy a slowly reconfigurable WDM technology

relying on commercially available ROADMs and

MEMS-based OXCs. This will lead to a significant

over-provisioning of links and large port-count

switches to accommodate dynamic traffic patterns.

In contrast, various dynamic resource allocation

schemes like optical burst/packet switching, provide

an alternative to over-provisioning but they

inherently lack QoS guarantees whilst they require

the deployment of rather immature optical

technology.

As proposed by A. Stavdas et al. (2008), an

alternative optical multiplexing scheme based on

TDM is CANON (Clustered Architecture for Nodes

in an Optical Network), that reconciles dynamic

resource allocation and statistical multiplexing gains

with QoS guarantees. The CANON architecture

refers to both a networking concept and the

corresponding switching solution.

In this work, we benchmark optical statistical

multiplexing schemes against static provisioning

over the existing infrastructure of the Pan-European

network topology as presented by S De Maesschalck

et al (2003) and when CANON architecture is

applied, to quantify performance results in terms of

loss performance and cost.

114

Drakos A., Orphanoudakis T., Politi C. and Stavdas A. (2010).

STATISTICAL TRAFFIC MULTIPLEXING WITH SERVICE GUARANTEES OVER OPTICAL CORE NETWORKS.

In Proceedings of the International Conference on Data Communication Networking and Optical Communication Systems, pages 114-118

DOI: 10.5220/0002991701140118

 SciTePress

2 ARCHITECTURAL

CONSIDERATIONS

In order to quantify the gains and trade-offs

comparing dynamic solutions vs. static provisioning

we first examine the case of wavelength switching

as a technology to implement the circuit switching

concept (Optical Circuit Switching - OCS). OCS

employs a pre-provisioned allocation of network

resources in order to serve the capacity demand.

OCS inherits both the advantages and disadvantages

of circuit switching: it guarantees QoS with no

losses or delay but at a high overall cost and poor

resource utilisation.

In order to overcome the severe scalability

limitations of OCS we consider sub-wavelength

reservation and switching techniques and evaluate

OBS as an indicative case. OBS, as proposed in the

literature, is a dynamic resource allocation protocol

and statistical multiplexing solution. An OBS node

aggregates traffic destined to one destination and

casts it into a burst after transmitting a reservation

message informing the intermediate nodes for the

upcoming burst transmission. It is well-known

through numerous studies, that OBS cannot

guarantee QoS due to this one-way reservation

scheme that leads to high burst losses. Slotted OBS

(S-OBS), presented by Z. Zhang, L. Liu and

Yuanyuan Yang (2007) is an OBS solution where all

bursts are constrained to a specific size. S-OBS is

considered a superior solution not only due to the

partial burst collision probability but also due to the

reduced switching node control and scheduling

complexity, when variable slot allocations need to

be scheduled in real-time. Therefore in this work,

the S-OBS is considered.

To limit the burst loss in large optical core

networks where dynamic resource allocation is

employed, CANON implements a hybrid reservation

mode. Electronic buffering is still employed at

network edge nodes, which are called Regular

Nodes (RNs) but optical frame generation is not

performed based on local RN queue status

information as is done in OBS nodes but based on a

MAC protocol executed between a Master Node

(MN) and all RNs of a sub-net called “cluster”. We

will briefly describe this principle of operation in the

remainder of this section and in the next section we

will demonstrate how this distributed switching

architecture leads to nearly optimal performance

results.

2.1 CANON Architecture

The CANON solution has previously described by

A. Stavdas et al. (2008) and by J. D. Angelopoulos

et al. (2007) and the principle of operation is now

summarized. CANON is organising the nodes of a

core network in clusters, mainly based on vicinity,

traffic, legacy infrastructure and administrative

criteria. This process creates a new hierarchy since

the core nodes are classified as Master Nodes (MN)

and Regular Nodes (RN) based on functionality

considerations and a differentiated traffic handling

policy (i.e. traffic destined to nodes of the same

cluster versus transit traffic destined to the nodes of

a distant cluster). The role of the MN is to

coordinate both inter and intra-cluster operations. In

the latter process, the RNs contribute fixed size

contiguous optical slots, which are marshalled into

appropriately sized frames destined for other

clusters, under the guidance of a reservation-based

MAC protocol. In addition, under MN’s supervision,

RNs use the same frame and wavelength to transport

traffic to the same destination cluster, so each ring is

effectively operating as a “distributed switch”.

Details on traffic aggregation, flexible bandwidth

allocation, robustness and resiliency are discussed

by J. D. Angelopoulos et al. (2007). Regarding inter-

cluster operation, a pre-provisioned scenario has

been analysed by J. D. Angelopoulos et al. (2007)

whilst a fully dynamic case based on one-way

reservations over the pan-European network has

been benchmarked by A. Stavdas, A.

Orphanoudakis, C. (T) Politi, A. Drakos and A. Lord

(2009).

Regarding CANON node architectures, the

proposed MN has been presented by A. Stavdas, C.

(T) Politi, T. Orphanoudakis and A. Drakos (2008).

It is a wavelength and a link modular architecture

allowing the node to gracefully scale to hundreds of

Tb/s. On the other hand, the RNs are optical

add/drop mulitplexers (OADMs). This solution

greatly supports the smooth migration from existing

rings to the CANON solution. Conclusively, the

combined operation of network and switch

architecture allows to efficiently groom slots into

frames in a collision free-way by means of a MAC.

3 BENCHMARKING CANON

3.1 Benchmarking Parameters

Since the performance of OBS is directly

proportional to the available number of wavelengths

STATISTICAL TRAFFIC MULTIPLEXING WITH SERVICE GUARANTEES OVER OPTICAL CORE NETWORKS

115

in a WDM system (since they provide the means for

contention resolution) we need to establish a fair

basis for comparison. The basic parameters for a

benchmarking between the different schemes

include the network topology, the available

resources and node functionality. Thus, we first

select a reference network topology and a traffic

load profile to evaluate network performance, when

serving this input traffic load. Obviously CANON

introduces a topology transformation and a different

multiplexing paradigm. Therefore the edge node

functionality in each case is different. However the

important thing we need to evaluate is the

performance of the resulting network architecture

given a specific amount of resources. In this study

the number of resources is expressed in terms of

number of ports and wavelengths per port on

network nodes. The basis for comparison is drawn

from the minimum number of wavelengths that are

required over the core network links to serve the

input traffic load in the static case of OCS. Given

this figure we demonstrate the performance trade-off

that sub-wavelength resource allocation can achieve

and benchmark the CANON distributed switching

architecture and OBS against static OCS.

For the benchmarking, the Pan-European core

topology of S De Maesschalck et al (2003) shown in

Figure 1 is used. This core network consists of 16

nodes interconnected in a partially mesh topology

consisting of 23 links. For our modelling work, the

following statistical characteristics of the incoming

traffic were assumed: Poisson inter-arrival times of

fixed-sized frames with size equal to 0.125msec and

uniform distribution probability destination. For

each node, traffic filling four wavelength channels

per input link was assumed. For the S-OBS case the

burst size has been assumed equal to 5msec. The

nodes are based on the λ-S-λ configuration presented

by A. Stavdas et al. (2008), providing full

wavelength conversion. Also at the edge nodes

electronic buffers have been assumed for all cases

where burstification takes place.

Firstly we study the performance of both S-OBS

and OCS. For the OCS scenario, each node needs to

have an interconnection with every other node and

taking into account the distribution of the four input

wavelength traffic load over all destinations and the

wavelength conversion capability at each output port

in every node, the minimum requirement for the

interconnection of the nodes is a (virtual) wavepath

of the size of one wavelength for each destination.

On the other hand for the S-OBS two different

scenarios have been evaluated. Since S-OBS is a

statistically multiplexing protocol with dynamic

reservations, it can be assumed that the over-

provisioning made at the OCS case can be limited to

cover only the expected demand as would be done in

any other case of traditional packet switching

networks. Thus the network capacity has been

limited to the average expected traffic of the OCS

capacity. In order to be more fair a second scenario

has been assumed. This time we adjust the capacity

of each link taking into account the maximum load

of a node. By using this over-provisioned

assumption we expect the performance of S-OBS to

be improved albeit with higher cost and poor

resource utilisation.

This difference in the number of resources of

each scenario and thus the cost of each solution can

also be seen in

Figure 2, with the two scenarios of S-

OBS marked as mesh S-OBS lim and mesh S-OBS

over respectively.

Figure 1: Pan-European core topology.

CANON

S-OBS

CANON

OCS

Mesh S-

OBS lim

Mesh S-

OBS

ove

Mesh

OCS

100

200

300

400

500

600

700

800

Capacity (num Tx/Rx pair)

Figure 2: Overall capacity in number of Tx/Rx pair.

In order to evaluate the benefits of CANON, we

segmented the same core network into 4 separate

clusters as it is indicated by the grey areas Figure 1

DCNET 2010 - International Conference on Data Communication Networking

116

without introducing any modification on the existing

infrastructure. The resulting topology is shown in

the inset of Figure 1 (upper right) showing 4 clusters

consisting from 4 MNs (nodes 3, 6, 7, 12) and a

number of RNs per cluster. Under the current

partitioning, those links which are designated with a

dotted line are not in use. It is important note that the

algorithm for the segmentation of the network

depends only on the distances of the nodes. A better

solution may be found by applying more efficient

algorithms, though such optimisation is out of the

scope of this paper.

By following the CANON solution, different

network capacities can be assumed for the two

hierarchical sections of intra and inter-cluster.

Additionally the traffic can also be distinguished in

traffic destined to other clusters, and traffic destined

to nodes inside the ring. For the latest, one

wavelength is assumed sufficient while for the rest

of the traffic the capacity of the ring is limited to the

required number in order to serve the average

expected traffic. For the inter-cluster, as before, the

two different approaches of OCS and S-OBS are

being used to serve the inter-cluster interconnection.

Following the same principle of the mesh OCS case,

in CANON-OCS, each MN needs a virtual wavepath

for each destination cluster but this time each

wavepath is provisioned only with the minimum

number of wavelengths required to serve the load.

Likewise, for the S-OBS case we limit the capacity

of each link to in order to serve the average expected

traffic taking also into account the routing.

In CANON, as explained earlier, only the MNs

are considered as λ-S-λ switches while the RNs are

OADMs. The same traffic profile as before is used

and the frames have the same duration of the S-OBS

burst of 5msec.

3.2 Result Evaluation

Figure 3 shows the loss probability of all simulated

scenarios. As expected no losses occur in the OCS

scenarios in contrast with the S-OBS. As it can been

seen in Figure 3 the mesh S-OBS solution with the

limited capacity even for the lowest traffic loads

suffers from network collisions and thus severe

losses. Moreover when the over-provisioned

solution for the mesh S-OBS network is used, the

performance improves but even for moderate

loading the loss probability still remains

unacceptably high (even though full wavelength

conversion is employed at all core nodes). On the

other hand when CANON is applied, a more reliable

system is shown and only for the highest traffic

loads there are losses that are more than an order of

magnitude less than what is achievable with mesh S-

OBS.

1,E-05

1,E-04

1,E-03

1,E-02

1,E-01

1,E+00

20% 40% 60% 80% 100%

Load (%)

Slot Loss Probability

Mesh S-OBS over

Mesh S-OBS lim

CANON S-OBS

* mesh OCS and CANON

OCS have 0% losses

Figure 3: Overall capacity in number of Tx/Rx pair.

Even though the simulations have a specific traffic

profile, in theory the maximum capacity of the

network would have been achieved only if all links

were loaded at 100% at the same time. Thus the

throughput of each scenario over this theoretical

maximum capacity of the network is a metric that

combines the results of both Figure 2 and Figure 3

and it is shown in Figure 4. Even though a mesh

OCS solution guarantees transmission QoS, it

demands an extremely high capacity which usually

remains under-utilized. On the other hand both of

the scenarios for the S-OBS case have poor

performance since they both suffer from severe

losses. Additionally in the case of the over-

provisioned network, the high cost reduces even

more the performance.

Finally, CANON is a solution that combines both

performance and cost. In the first case the loss

probability of CANON S-OBS is close to the

guaranteed performance of OCS due to the statistical

multiplexing enforced by CANON, while even when

OCS is used for the inter-cluster communication, the

new architecture enforced by CANON keeps the

demanded network resources close to the values of

the S-OBS case. Thus, as shown in Figure 4, both

CANON solutions have similar performance.

Moreover both CANON solutions not only

outperform the mesh scenarios and if we also take

into account the different switching technologies

that CANON uses in the intra-cluster segment we

can also deduce that the overall cost of CANON

(monetary cost, power consumption,…) is much

lower than the compared solutions.

STATISTICAL TRAFFIC MULTIPLEXING WITH SERVICE GUARANTEES OVER OPTICAL CORE NETWORKS

117

0,05

0,1

0,15

0,2

0,25

20% 40% 60% 80% 100%

Load (%)

Throughput/Capacity

Mesh S-OBS over

Mesh OCS

CANON OCS and

CANON S-OBS

Mesh S-OBS

lim

Figure 4: Throughput over theoretical maximum capacity.

4 CONCLUSIONS

We demonstrated that using the existing

infrastructure of the Pan-European network,

CANON can efficiently combine statistical

multiplexing gains and improved blocking

performance, compared to a mesh S-OBS solution,

while keeping the overall cost of the implementation

considerably lower than the pre-provisioned mesh

OCS case. S-OBS suffers from extensive loss

probability resulting in almost complete blocking

even at light traffic load values and even under

resource over-provisioning. CANON on the other

hand can provide the same level of QoS guarantees

as OCS (practically lossless operation), limiting

over-provisioning by employing the OCS model

only in the inter-cluster network and utilizing

statistical multiplexing that results from the

distributed traffic aggregation inside the local

network clusters. Even in the fully dynamic case

when OBS is employed for MN interconnection over

the mesh inter-cluster topology, the cost (in terms of

resources) and performance trade-offs are much

better than in the case of S-OBS.

ACKNOWLEDGEMENTS

The work described in this article was carried out

with the support of the STRONGEST-project funded

by the European Commission through the 7th ICT-

Framework Programme.

REFERENCES

A. Stavdas, T. G. Orphanoudakis, A. Lord, H. C. Leligou,

K. Kanonakis, C. Matrakidis, A. Drakos, J. D.

Angelopoulos, “Dynamic CANON: A Scalable Multi-

domain Core Network”, IEEE Communications

Magazine, Special Issue on Multi-Domain Optical

Networks: Issues and Challenges, Vol. 46, Issue 6, pp.

138-144, June 2008.

S De Maesschalck et Al, “Pan-European Optical Transport

Networks: An Availability-based Comparison” ,

Photonic Network Communications, Vol. 5, No. 3, pp.

203-225, May 2003.

J. D. Angelopoulos, K. Kanonakis, G. Koukouvakis, H.C.

Leligou, C. Matrakidis, Th. Orphanoudakis and A.

Stavdas, “An Optical Network Architecture with

Distributed Switching Inside Node Clusters Features

Improved Loss, Efficiency and Cost”, IEEE Journal of

Lightwave Technology, Vol. 25, No. 5, May 2007.

A. Stavdas, A. Orphanoudakis, C. (T). Politi, A. Drakos,

A. Lord , “Design, performance evaluation and energy

efficiency of optical core networks based on the canon

architecture” IEEE/OSA Optical Fiber

Communication Conference and Exposition (OFC),

San Diego, CA, USA, March 22–26 2009

A. Stavdas, C. (T) Politi, T. Orphanoudakis, and A.

Drakos, “Optical packet routers: how they can

efficiently and cost-effectively scale to petabits per

second”, OSA Journal of Optical Networking, Vol.7,

No. 10, pp. 876-894, 2008

Z. Zhang, L. Liu and Yuanyuan Yang, “Slotted Optical

Burst Switching (SOBS) networks”, Computer

Communications, vol. 30, no. 18, Dec. 2007, pp. 3471-

3479.

DCNET 2010 - International Conference on Data Communication Networking

118