RETHINKING ISP ARCHITECTURES THROUGH

VIRTUALIZATION

Alejandro Cordero and David Fernández

Departamento de Ingeniería de Sistemas Telemáticos, Universidad Politécnica de Madrid, Madrid, Spain

Keywords: Network virtualization, ISP architectures, Future internet.

Abstract: Network virtualization is a powerful technology that has been proposed as a mean to evolve the current

Internet, allowing the introduction and testing of future network technologies over present infrastructures in

coexistence with our current production networks. However, apart from a path to the Future Internet,

network virtualization can also be a key technology that modifies and improves today ISP networks. This

paper analyses how network virtualization and all the technologies being developed around it can influence

and evolve the present ISP network architectures and business models. Starting from the well-known ISP

architecture made of access, distribution and core layer; we present and discuss the benefits that can be

achieved by introducing virtualization technologies in each layer.

1 INTRODUCTION

Virtualization is a mature technology in the context

of computing resources, mainly in what is related to

operating systems (platform virtualization). Many

data centres nowadays are supported by

virtualization suites that allow high flexibility in

sharing physical machines and provide capabilities

like high performance, availability, energy

efficiency, load balancing, etc.

However, in the context of networking,

virtualization is a recent technology. Network

virtualization consists of creating multiple logical

networks over a common physical infrastructure

made of nodes and links. Physical nodes are shared

using platform virtualization techniques adapted to

communication nodes, taking into account their

strict real-time requirements, as well as the

reliability (carrier-class).

Network virtualization has been proposed as a

mean to allow the coexistence over the same

network infrastructure of the current Internet and the

new network architectures and protocols being

proposed for the Future Internet. In fact, several

research programs and projects are working in the

application of network virtualization to Future

Internet (e.g. GENI in USA, FIRE in EU or AKARI

in Japan). But it has also a clear role to play in the

evolution of current Internet architectures, creating

new roles and business models. All the new

technologies developed around virtualization, like

router virtualization or dynamic reconfiguration of

networks can be applied to current network

architectures, giving rise to more flexible and

efficient networks.

This paper analyses how network virtualization

and related technologies can influence and evolve

the present ISP network architectures and business

models. The paper is organized as follows. Starting

from the well-known ISP architecture (Section 2)

made of access (2.1), distribution (2.2) and core

(2.3) layers; we present and discuss the benefits that

can be achieved by introducing these virtualization

technologies in each layer. Finally, we provide some

concluding remarks in Section 3.

2 VIRTUALIZING ISP

ARCHITECTURE

In order to organize our discussion about how

network virtualization can be applied to today ISP

networks, in this section we will briefly describe the

three-layer network hierarchical model

(Oppenheimer, 2004), a highly adopted “de facto”

standard for ISP (and Enterprise) network design

topologies. This model permits traffic aggregation

and filtering at three successive routing or switching

Cordero A. and Fernández D..

RETHINKING ISP ARCHITECTURES THROUGH VIRTUALIZATION.

DOI: 10.5220/0003443400810084

In Proceedings of the International Conference on Data Communication Networking and Optical Communication System (DCNET-2011), pages 81-84

ISBN: 978-989-8425-69-0

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

levels and makes it very scalable up to large

international internetworks levels.

A typical hierarchical network topology consists

in:

 A core layer of high-end routers and switches

that are optimized for availability and

performance.

 A distribution layer of routers and switches

that implement policies.

 An access layer that connects users via lower-

end switches and wireless access points.

The next subsections provide some insights

about how network virtualization can be applied to

each of the layers described before.

2.1 Access Layer

The access layer provides users on local segments

the access to the internetwork. The access layer can

include a large variety of equipment, such as routers,

switches, bridges, shared-media hubs, wireless

access points, etc., using different technologies such

as xDSL, FTTH, Ethernet, DOCSIS, UMTS/3G, Wi-

Fi, WiMAX, etc, as well as different physical media

such as copper (twisted pair and coaxial), fiber and

air.

From the point of view of traditional ISP, we

divide the access layer in three differentiated

components for analyze their virtualization

possibilities:

 The Customer Equipment.

 The Provider Equipment.

Customer Equipment: The virtualization of

customer equipment like residential or home

gateways is analyzed in (Royon, Frénot, 2007),

(Ibanez et al., 2007). OSGi is a common solution to

the CE implementation that permits a fast an easy

deployment of new services and protocols.

Virtualization of OSGi can be seen as a method for

sharing CE among providers.

Another result of the virtualization of CE is

decoupling service from devices that permits the

creation of new solutions. i.e. a Virtual Home

Environment (Berl et al., 2009) architecture created

to move and consolidate services to ensure energy

efficiency, or a service virtualizer (Häber et al.,

2009) where remote devices are presented as if they

were in local network.

However, we must not forget the hardware

needed to achieve an optimal implementation of a

CE with virtualization capabilities. Today residential

gateways are focused to Internet access as well as its

hardware. Future implementation of a CE must be

more flexible to allow joining several services in one

device. This flexibility should permit upgrading to

new protocols and implementing new services

without changing equipment.

The key of success for this approach depends on

the equipment cost versus the flexibility that it

grants, but the possibility of sharing CE cost

between providers improves the chance to win.

Therefore, from a user point of view, virtualization

can improve customer equipment allowing:

 Faster deployment of new services.

 Easier implementation of new protocols.

 Sharing CE between providers.

 Consolidating equipment to improve energy

efficiency (green compliance).

Provider Equipment: Today’s ISPs have link

layer equipment such as DSLAM, CMTS, ONT or

RNC, depending on the transmission technology

used. This equipment could be virtualized to allow it

to be shared among several VNP or to be aggregated

into one to facilitate its management. In this case, it

could be possible to easily migrate a user inside the

same equipment between VNP or upgrade link layer

protocols, but we always have the limitation

imposed by physical end-user connection.

Currently, ISP’s solve the problem of access

layer in two ways:

 Deploying their own access network.

 Paying traffic aggregation of their users to its

distribution layer.

Virtualization would allow the deployment of an

intermediate model, where an ISP with its own

network could become the infrastructure provider of

other ISP’s behaving as virtual network providers.

This model has two direct consequences:

 The VNP has greater control over traffic at

limited cost.

 The IP could share its resources and have a

new business source. This IP focused on the

virtualization of the access layer could be

called network access operator (NAO).

Another consequence of virtualization is the

faster creation of new VNP networks, where

deployment times could be significantly reduced.

Moreover, depending on the level of virtualization

of the devices, a VNP could implement different

versions of the protocols in their virtual instance, but

this would imply more resources and added

complexity.

However, access technologies are not prepared

for virtualization. Today vendors only have done

some advances in Ethernet switching. FlowVisor

(Sherwood et al., 2009), is a special purpose

OpenFlow controller that acts as a transparent proxy

between OpenFlow switches and multiple

DCNET 2011 - International Conference on Data Communication Networking

controllers. This solution may have application for

today’s MetroEthernet access networks, dividing it

in a better way than traditional VLANs, but in order

to have enough capabilities for VNP, future

implementations need to have a separate control and

configuration for each virtual instance.

Therefore, from the provider point of view, the

use of virtualization has several benefits:

 More network control like local loop

unbundling implementations.

 Easier implementation of new solutions.

 Sharing cost between providers.

2.2 Distribution Layer

The distribution layer of the network is the

demarcation point between the access and core

layers. The distribution layer controls the access to

resources and network traffic for security and

performance reasons, hiding detailed topology

information about the access layer from core routers.

Also, the distribution layer allows the core layer to

connect access layers that run different protocols or

technologies while maintaining high performance. In

summary, its key needs are traffic optimization,

security, and media transitions.

Network virtualization can improve distribution

layer in the same way as shown in the access layer

reducing costs by sharing equipment and providing a

better way to deploy new solutions and protocols.

But in the current implementations of the

distribution layer in ISP’s, mostly based in MPLS

technologies, other improvements can be derived

from the virtualization of MPLS PE (Figure 1).

Figure 1: Virtual PE at the distribution layer.

The idea of a virtual PE (vPE) is a logic step in

the evolution of virtual private networks, where

configurations and traffic are better isolated,

increasing security levels. Moreover, virtual PE

implementation gives rise to the possibility of router

migration, a similar functionality to the virtual

machines migration found in virtual computing

platforms but applied to routers. In VROOM (Wang

et al., 2008) router live migration is analyzed as a

management primitive for planned maintenance and

service deployment. Virtual PE consolidation (i.e.

grouping several vPE’s over one real node) could be

another benefit, minimizing power consumption

through the hibernation of unused real equipment.

However, new developments are needed in order

to deploy new capabilities like programmability for

new protocol deployment or mobility primitives.

The migration of virtual routers capability opens

the possibility to use dynamic reallocation

techniques in the distribution layer. This possibility

will be analyzed deeper in next section applied to

core layer.

Therefore, network virtualization can improve

distribution layer in the following way:

 Improving security through isolation of clients

and processes.

 Migrating virtual routers to aggregate access

layer in less physical nodes according to

network traffic.

 Hibernating portions of distribution layer to

optimize power consumption, and ensure

green energy compliance.

2.3 Core Layer

The core layer is the high-speed backbone of the

internetwork. The core should have a limited and

consistent diameter, designed with redundant

components, in order to be highly reliable and to

adapt to changes quickly. Moreover, the core layer

needs to be optimized for low latency and packet

throughput.

The creation of a Composable Router with

virtualization could be a good solution to implement

scalable megarouters reducing costs. With this

approach we could have high speed core routers by

reusing common routers. The basic idea consists on

separating control and data planes: in the data plane,

composed of an array of routers, traffic is managed

by and splitter that realises distribution and

aggregation features; in the control plane,

management is done by a meta-router that analyses

control packets and creates and maintains

forwarding tables.

Today ISP networks use redundancy in its design

as a solution to failure, creating complex

architectures to operate and support. Virtualization

can change the way that today designs are made,

dividing network design between IP and VNP. If

redundancy is provided by IPs, deploying a physical

redundant substrate with dynamic reallocation

capabilities triggered by network failures or resource

depletion, network design will be simplified for

VNP.

Therefore, a well-designed IP network would

allow simple fault tolerant VNPs networks,

RETHINKING ISP ARCHITECTURES THROUGH VIRTUALIZATION

improving scalability and optimizing resource use

while minimizing cost. But, how can this be done?

Different approaches can be implemented to solve

the network-embedding problem, like restricting the

problem space (sacrificing flexibility) or proposing

heuristic algorithms (sacrificing efficiency). Some of

these approaches are taken in (Zhu, Ammar, 2006)

(Fan, Ammar, 2006) (Yu et al., 2008), but finding

the optimal solution turns out to be a NP-hard

problem.

However, today core architectures, composed by

simple topologies with a limited number of devices,

could permit the implementation of dynamic

reallocation. This controlled approach can simplify

the NP-hard problem, providing a way to create fully

meshed substrate networks based on core networks

topologies.

Finally, in core layer, as in all layers, the use of

shared equipment among providers is an option. In

this case, virtual cores can be created to connect

isolated networks, having cheaper and self-managed

high-speed transport.

Therefore, network virtualization can improve

core layer in the following ways:

 Creating virtual routers that aggregate several

real ones. This facilitates management having

only one interface for all devices. Also offers

scalability on demand, adding devices when

more power is needed.

 Implementing virtual dynamic reallocation

support fault tolerant networks. This

simplifies VNP networks design over

redundant IP substrate networks.

3 CONCLUSIONS

Virtualization is a mature technique in the

computing environment that applied to networks can

bring new solutions and business opportunities to the

actors involved in today Internet. The first steps in

this field have been already taken by research

programs and industry vendors, creating an

environment for future development. However, there

is still much work to be done.

On the other hand, although ISP architectures are

very consolidated, virtualization can improve their

design providing new possibilities not seen before.

In this paper we have presented an analysis about the

benefits that network virtualization can bring to each

of the layers of an ISP architecture. They can be

summarized as follows:

 Consolidation of equipment to improve energy

efficiency.

 Faster and easier implementation of solutions

and protocols.

 Sharing cost between providers.

Finally, we believe that network virtualization

techniques can be implemented in a near future,

improving today’s ISP networks. However, more

insights into the possibilities outlined in this article

are needed for a virtual networking environment to

become a reality.

REFERENCES

Berl, A., Weidlich, R., Schrank, M., Hlavacs, H. & de

Meer, H. 2009, "Network Virtualization in Future

Home Environments", Integrated Management of

Systems, Services, Processes and People in IT, pp.

177-190.

Fan, J. & Ammar, M. 2006, "Dynamic topology

configuration in service overlay networks: A study of

reconfiguration policies", Proc. IEEE

INFOCOMCiteseer.

Häber, A., De Mier, J.G.R. & Reichert, F. 2009,

"Virtualization of Remote Devices and Services in

Residential Networks", 2009 Third International

Conference on Next Generation Mobile Applications,

Services and TechnologiesIEEE, , pp. 182.

Ibanez, M., Madrid, N.M. & Seepold, R. 2007,

"Virtualization of Residential Gateways", Intelligent

Solutions in Embedded Systems, 2007 Fifth Workshop

onIEEE, pp. 115.

Oppenheimer, P. 2004, Top-down network design, Cisco

Press.

Royon, Y. & Frénot, S. 2007, "Multiservice home

gateways: Business model, execution environment,

management infrastructure", IEEE Communications

Magazine, vol. 45, no. 10, pp. 122-128.

Sherwood, R., Gibb, G., Yap, K.K., Appenzeller, G.,

Casado, M., McKeown, N. & Parulkar, G. 2009,

"Flowvisor: A network virtualization layer".

Wang, Y., Keller, E., Biskeborn, B., van der Merwe, J. &

Rexford, J. 2008, "Virtual routers on the move: live

router migration as a network-management primitive",

ACM SIGCOMM Computer Communication Review,

vol. 38, no. 4, pp. 231-242.

Yu, M., Yi, Y., Rexford, J. & Chiang, M. 2008,

"Rethinking virtual network embedding: Substrate

support for path splitting and migration", Computer

Communication Review, vol. 38, no. 2, pp. 17.

Zhu, Y. & Ammar, M. 2006, "Algorithms for assigning

substrate network resources to virtual network

components", Proc. IEEE INFOCOMCiteseer.

DCNET 2011 - International Conference on Data Communication Networking