The Internet has stopped evolving, but how can we start the evolution up again? We have already seen that a new technology, that wants to have any hope of actually being deployed, has to have two properties. It needs to be backwards compatible and incrementally deployable. Network virtualization has been put forward as a suitable candidate [6, 12, 66, 168, 171, 172].
The basic idea of network virtualization is straight forward. Instead of using one physical work that can do everything well, use multiple virtual networks embedded in the physical net-work, each one specialized and perfectly adapted for a particular application. To allow for this adaptation, the nodes of the virtual networks receive compute capabilities within the routers of the physical network. Therefore, virtual networks can offer their own (and application specific) topology, routing, naming services, and resource management [171]. The alternative to adapted virtual networks, multiple physical networks, is clearly infeasible on a global scale.
That sounds promising, but is it realistic? At the very least, network virtualization requires sup-port from the physical network by offering routers that can host virtual machines (the routers of the virtual networks). This technology is already available [32]. In addition, network vir-tualization is actively and successfully used in large scale scientific network testbeds such as GENI [62], PlanetLab [31] or G-Lab [156], not as an enhancement to be studied, but as a central enabling technology for carrying out experiments. Virtual networks are used to partition the network testbeds so that different research groups can perform their experiments without inter-ference from each other. As Tutschku et al. [172] state, the “virtualization of telecommunication services or applications is no longer an academic concept”.
Network virtualization is not the only proposed concept to break the ossification of the Internet.
Alternatives include OpenFlow [123] and Software Defined Networks [112], which allow very flexible routing that can be administrated centrally.
1.3.1 Advantages and Applications
Allowing network virtualization in the Internet offers some advantages and applications, which are currently unthinkable, become possible. As a central advantage, Berl et al. [12] mention the flexibility of the system. Virtual networks can be dynamically reconfigured, new networks can be added and old ones can be removed, suspended or discarded. The state of the network can be frozen and reverted if the need should arise. Unused parts of the network can be shut down to conserve energy. In aggregated services, it is possible to fix or replace parts transparently [27].
Some interesting possibilities also emerge on the user side of things. Turner et al. [170] describe a virtual network offering a learning environment with high quality audio and video multicast mechanisms. Format translators are available at the virtual nodes to enhance compatibility. With virtual networks, it may be possible to switch Internet Service Providers (ISPs) on the spot, like it is possible for electricity or phone providers [171].
1.3.2 Challenges
Adoption is a central problem for every new technology. How will (or should) the adoption process of network virtualization work? Anderson et al. [6] envision the following adoption process. Virtual networking will start with a single daring Next Generation Service Provider (NGSP) offering virtualization services for its own network. Customers not directly connected to the network of the NGSP can connect via standard ways through the current Internet. If the NGSP is successful, it can expand its network to reach more customers. Local ISPs may be forced to offer the same services to stay competitive. For Turner et al. [170], adoption will resemble more the introduction of the Internet. First, virtual networks will be offered as an over-lay in the existing Internet. Then, a government-supported experimental backbone infrastructure will be built, which natively supports virtualization. As the last step follows the commercial op-eration of virtual networks.
After adoption is achieved, the tasks of the current ISPs will be carried out by two differ-ent business differ-entities [29], the infrastructure providers (InPs) and the virtual network providers (VNPs). InPs will manage the physical infrastructure necessary for hosting the virtual
net-works. VNPs create virtual networks from a federation of the resources offered by the different InPs [28, 50, 170].
This structure offers a rich environment for business opportunities [170]. InPs can compete by offering better services for VNPs using their networks, such as high quality resources, manage-ment tools, operation support, and fault tolerance. VNPs can distinguish themselves by offering shorter virtual network setup times, higher quality virtual networks, or guaranteed resources. In addition, VNPs do not own any physical resources, they do not need to deploy and maintain infrastructure. Thus, there is a low barrier to entry for VNPs. There are also opportunities for network equipment vendors, as there will be demand for high performance virtualizable routers.
Network virtualization is not ready to be deployed in the Internet. The authors of [29, 170]
identify key research questions that still need to be answered, which we summarize below.
Interfacing How can InPs and VNPs communicate, for instance about available resources or requirements?
Signaling and Bootstrapping How can VNPs set up their virtual resources, if they have no communication capabilities besides the resources that they want to set up?
Resource Allocation How can a VNP best fit its virtual networks into the resources it has leased from the InPs?
Resource Discovery How can InPs keep track of the resources they offer and their connectivity, especially to other InPs.
Admission Control How can it be ensured that the capacities of an InP cannot be exceeded?
Virtualization How are the physical routers to be designed to allow low overhead virtualiza-tion?
Resource Scheduling How can an InP efficiently distribute its resources among the interested VNPs. How long should an InP guarantee the availability of resources?
Naming and Addressing How can the situation be handled that a single host may connect to multiple different virtual networks, each with different naming and addressing schemes?
Dynamism and Mobility How can the dynamic nature of virtual networks and the changing location of users be efficiently handled in terms of routing?
Operation and Management How can virtual networks be efficiently monitored and man-aged?
Security and Privacy How can it be prevented that a hostile virtual network breaks out of its virtualized environment and takes control of the physical infrastructure?
Heterogeneity of Technology How can the plurality of different virtualization technologies be handled efficiently?
Economics How should the economics of virtual networks work?
Service Duplication How can the overhead caused by multiple virtual networks offering the same basic services be avoided?
In this thesis, we cannot solve all those problems. In the following section, we will outline our aim.