Results (
Vietnamese) 2:
[Copy]Copied!
DESIGN GOALS
The design goals for successfully realizing net- work virtualization have been addressed by dif- ferent research groups. In order to materialize network virtualization, each of these design cri- teria should be fulfilled.
Flexibility — Network virtualization must pro- vide freedom in every aspect of networking. Each SP should be free to implement arbitrary network topology, routing and forwarding func- tionalities, and customized control protocols independent of the underlying physical network and other coexisting VNs. For example, deploy- ing source routing in today’s network depends much on consensus among ISPs; in a virtualized environment, the owner of a VN should be able to offer source routing without having to coordi- nate with any other parties.
Manageability — By separating SPs from InPs, network virtualization will modularize network management tasks and introduce accountability at every layer of networking [3]. It must provide complete end-to-end control of the VNs to the SPs, obviating the requirement of coordination across administrative boundaries seen in the existing Internet.
Scalability — Coexistence of multiple networks is one of the fundamental principles of network virtualization. Scalability is an indispensable part of this equation. InPs in an NVE must scale to support an increasing number of coexisting VNs without affecting their performance.
Isolation — Network virtualization must ensure isolation between coexisting VNs to improve fault tolerance, security, and privacy. Network proto- cols are prone to misconfigurations and imple- mentation errors. Virtualization must ensure that misconfigurations in one VN are contained within itself and do not affect other coexisting VNs.
Stability and Convergence — Isolation ensures that faults in one VN do not affect other coexisting VNs, but errors and misconfigurations in the underlying physical network can also destabilize an NVE. Moreover, instability in the InPs (e.g., routing oscillation) can lead to insta- bility of all hosted VNs. Virtualization must ensure the stability of an NVE, and in case of any instability the affected VNs must be able to successfully converge to their stable states.
Programmability — To ensure flexibility and manageability, programmability of the network elements is an indispensable requirement. Only through programmability can SPs implement customized protocols and deploy diverse ser- vices. Two pressing questions in this respect must have satisfactory answers: “How much pro- grammability should be allowed?” and “How
should it be exposed?” A win-win situation must be found where programmability is easy, effec- tive, and secure at the same time.
Heterogeneity — Heterogeneity in the context of network virtualization comes mainly from two fronts: first, heterogeneity of the underlying net- working technologies (e.g., optical, wireless, and sensor); second, each end-to-end VN, created on top of that heterogeneous combination of underly- ing networks, can also be heterogeneous. SPs must be allowed to compose and run cross-domain end-to-end VNs without the need for any technology- specific solutions. Underlying infrastructures must also be capable of supporting heterogeneous pro- tocols and algorithms implemented by different SPs. In addition, heterogeneity of end-user devices must also be taken into account.
Legacy Support — Legacy support or back- ward compatibility has always been a matter of deep concern while deploying any new technolo- gy. Conceptually, network virtualization can easi- ly integrate legacy support by considering the existing Internet as just another VN in its collec- tion of networks; but whether and how it can be done efficiently remains an open challenge.
NETWORK VIRTUALIZATION
PROJECTS
Over the years, the term virtual network has been used to describe different projects on VPNs, overlay networks, and active or programmable networks. But very few of them actually followed the pluralist view of network virtualization. In Table 1 we summarize the most significant past and on-goingprojects directly or indirectly relat- ed to network virtualization based on the follow- ing set of characteristics:
•Networking technology: A handful of network virtualization prototypes have been devel- oped for specific networking technologies with an aim to exploit unique characteristics of those networks to enable virtualization. Such projects include X-Bone for IP net- works, Tempest targeting ATM networks, and the very recent GENI initiative that will be agnostic to any specific technology.
•Layer of virtualization: Influenced by the existing Internet, researchers have naturally approached network virtualization in a lay- ered manner. As a result, many projects have attempted to virtualize different layers of the network stack, starting from the physical layer (UCLP) and continuing up to the application layer (VIOLIN).
•Architectural domain: Most projects have focused on particular architectural domains, which dictate the design choices made in the construction of architectures and services that can be offered on those platforms. Examples include network management (VNRMS), vir- tual active networks (NetScript), and spawn- ing networks (Genesis).
•Level of virtualization: To enable network virtualization, one must virtualize the nodes, links, and every other resource in the net- work. The level of virtualization refers to
Network virtualiza- tion must provide freedom at every aspect of network- ing. Each SP should be free to implement
arbitrary network topology, routing and forwarding func- tionalities, and cus- tomized control protocols indepen- dent of the underlying physical network and other coexisting VNs.
Being translated, please wait..
