Network Virtualization: a path towards Internet innovation?

Author: Elio Salvadori, Roberto Doriguzzi (CREATE-NET)
Email: name.surname@create-net.org

Abstract: While current Internet seems to face a dramatic crisis due to the extreme ossifications induced by the current world-wide spread architecture, the scientific community is lately pushing toward the adoption of network virtualization techniques to overcome this impasse. In spite of the amount of open-issues which must be properly solved on the research side yet, initial results obtained from preliminary projects on network virtualization are confirming the strong potentiality of this technology. The objective of this article is to emphasize the impact of network virtualization as a tool to foster innovation both on the service and on network infrastructures side.

Internet has been proving to be one of the most successful revolution human society has ever been through. The architecture initially thought to realize a reliable communication network infrastructure for military scopes has been through a constant technological evolution as soon as its pervasive usage changed its original aim to a more commercial extent. However, the world-wide adoption of Internet TCP/IP-based architecture combined with its intrinsic multi-provider nature raised some major implications: modifications to this network architecture to accommodate new services or to manage billions of devices in the next future are - if not impossible - extremely difficult to realize due to the very large consensus needed among current stakeholders and to the potentially dangerous impact on services in operation. The introduction of any innovation on the existing architecture is therefore extremely difficult to realize, even though the impelling problems are severely defying current Internet architecture.

In recent years, a new approach to cope with this imminent problem of the Future Internet has emerged from the scientific community: network virtualization. As any known computer virtualization technology, network virtualization is based on the basic concept of abstracting a real resource by obscuring its underlying complexity. The basic idea is to use network virtualization as enabling technique to run several instances of network protocols/architectures each on a separate slice of a shared physical infrastructure; by guaranteeing the coexistence between legacy and innovative architectures, network virtualization facilitates a smooth swap toward novel architectural solutions on a production network. However, while products for computer and server virtualization are already in their mature state, with tens of commercial products already available on the market, no clear path can be envisioned yet on network virtualization techniques to be adopted inside current world-wide Internet architecture.

Even though several research initiatives on network virtualization are on-going currently, either in US, Europe and Japan, there's no common definition of what this technique refers to exactly. This is partially due to historical reasons behind what "virtualization" has meant in networks so far but also because several approaches with similar aims are currently available from the research community, each generally focusing on a specific layer of application of this technique at network architecture level. While on one side network virtualization should not be confused with existing technology like Virtual Private Networks, Active/Programmable Networks or Overlay Networks, on the other side recent approaches to this technique inherit some of the features of all these architectural solutions.

For example, some of the research projects have been working toward introducing virtualization concepts into specific layers of the networking stack, like UCLP [1] or Orbit [2] on the physical layer, AGAVE [3] or OpenFlow [4] on the network layer, VIOLIN [5] on the application layer. However for all these projects, the substrate technology is generally fixed (either IP, SONET or Ethernet) therefore limiting the real impact of this technique. Other research projects have been instead working toward the virtualization of network resources themselves; a pioneering project is PlanetLab [6] which proposes an overlay-based testbed to create virtual networks realized by interconnecting virtual machines on different nodes. Its extension VINI [7] guarantees further flexibility to researchers who wanted to test novel routing protocols by allocating "slices" of a physical infrastructure to several concurrent virtual networks. Lessons learned in PlanetLab have been fostering the set up of more ambitious initiatives worldwide like GENI [8] in US, AKARI [9] in Japan and FIRE [10] in Europe, where network virtualization techniques play a key role in overcoming the so-called "Internet impasse".

Even though network virtualization techniques have been mainly investigated by the international research community to provide controlled experimentation of new network technologies to solve current Internet ossification, several innovative applications leveraging on these techniques can be envisioned and are briefly described in the following.

One of the most important impacts of network virtualization on current Internet scenario refers to the possibility to support the coexistence of multiple heterogeneous network architectures from different service providers which are sharing a common physical substrate operated by a number of infrastructure providers. A trend in this direction can be already perceived in today market, e.g. the Spanish WISP FON is leveraging on third-party physical infrastructure to provide "global" Internet access through their low-cost Access Points. Network virtualization can be seen as a tool to guarantee a clear decoupling between service and infrastructure providers, a key property to guarantee innovation in current Internet. One of the most evident advantages of such an environment is the possibility to lower the cost of the physical substrate by sharing CAPEX and OPEX among several infrastructure providers. A network virtualization model generalizing the concept of New-Generation Service Provider (NGSP) originally identified in [11] has been recently proposed [12,13]. According to this approach, four entities can be identified: (i) Infrastructure providers, responsible for deploying and operating the physical resources as well as for providing clear access to the infrastructure to their customers. Multiple Infrastructure providers can collaborate through specific agreements in order to provide the widest substrate to Connectivity providers. (ii) Connectivity providers create multiple virtual networks over physical resources leased from Infrastructure providers; they are the entities aggregating substrates from different Infrastructure providers (i.e. access and transport physical networks) to match Service providers' requirements to available physical resources. (iii) Service providers can deploy new network protocols over these virtual networks to provide advanced services to their users, while of course concurrently furnishing traditional TCP/IP based services as well. (iv) End users access to multiple services provided by several Service providers.

All the major initiatives worldwide (GENI, AKARI and FEDERICA [14]) are actually pushing this vision behind the wide term "network virtualization". Their main objective is to solve some key open-issues such as flexibility, manageability, scalability, isolation, security, heterogeneity, etc... on a medium-term basis. A demonstration of the advantages of such vision can be already perceived from the initial results obtained out of smaller projects like CABO [15] or VROOM [16].

One of the most evident reward for providers applying network virtualization techniques is in term of improved operation and maintenance of their (virtual or physical) network infrastructure. For example, instantiating a virtual network based on network protocols tailored over specific services can guarantee the delivery of high-quality and trusted services to end-users. An ISP instantiating virtual nodes on remote sites of a third infrastructure provider can minimize the costs of deploying and operating these equipments on his own. Performing test of new protocols on a safe network slice sharing the physical infrastructure of the network slice in production guarantees more reliable results than canonical test activities performed in a lab or on a geographically limited testbed. Running several virtual network instances on the same infrastructure guarantees limited or no impact on existing configurations of a production network when new services must be provided to end users, by leveraging on the isolation guaranteed between each instance.

Another interesting application of these techniques refers to the possibility to move virtual routers between different physical locations inside the network; while simplifying existing network maintenance tasks, this application can be seen as a tool to reduce energy consumption in the network, a rising challenge nowadays.

In conclusion, network virtualization is going to play a key role in Future Internet not only as enabler for the development of new network protocols and architectures but also as a tool for introducing innovation into current worldwide Internet Service Provider scenarios, by effectively triggering a decoupling between infrastructure and service providers and by improving the operation and maintenance of their networks.

References:

[1] UCLP (User Controlled Lightpaths) - http://www.uclp.ca/

[2] Orbit project at University of Rutgers, Columbia - http://www.orbit-lab.org/

[3] AGAVE, FP6 EU project, http://www.ist-agave.org

[4] The OpenFlow Switch Consortium, http://www.openflowswitch.org

[5] Xuxian Jiang, Dongyan Xu, "VIOLIN: Virtual Internetworking on OverLay INfrastructure", Purdue University, Department of Computer Sciences, Technical Report CSD TR 03-027, 2003

[6] PlanetLab, http://www.planet-lab.org/

[7] VINI, http://www.vini-veritas.net/

[8] GENI, http://www.geni.net/

[9] AKARI, http://akari-project.nict.go.jp/

[10] FIRE, http://cordis.europa.eu/fp7/ict/fire/

[11] Scott Shenker, Larry Peterson and Jonathan Turner. "Overcoming the Internet Impasse through Virtualization, Proceedings of Hotnets, November 2004

[12] New Yaping Zhu, Rui Zhang-Shen, Sampath Rangarajan, and Jennifer Rexford, "Cabernet: Connectivity architecture for better network services," to appear in Proc. Workshop on Rearchitecting the Internet, December 2008

[13] N.M. Mosharaf Kabir Chowdhury and Raouf Boutaba, "A Survey of Network Virtualization", University of Waterloo, Ontario, Canada, Technical Report CS-2008-25, October 2008

[14] FEDERICA, FP7 EU project, http://www.fp7-federica.eu/

[15] Nick Feamster, Lixin Gao, and Jennifer Rexford, "How to lease the Internet in your spare time," Georgia Tech, Technical Report GT-CSS-06-10, August 2006

[16] Yi Wang, Eric Keller, Brian Biskeborn, Jacobus van der Merwe, and Jennifer Rexford, "Virtual routers on the move: Live router migration as a network-management primitive," in Proc. ACM SIGCOMM, August 2008