Results (
Vietnamese) 2:
[Copy]Copied!
5.7. Resource scheduling
When establishing a virtual network, a service provider requires specific guarantees for the virtual nodes’ attributes as well as the virtual links’ bandwidth allocated to its network [3]. For virtual routers, a service provider might request guarantees for a minimum packet processing rate of the CPU, specific disk requirements, and a lower bound on the size of the memory. On the other hand, virtual link requests may range from best-effort service to fixed loss and delay characteristics found in dedicated physical links. To provide such guarantees and to create an illusion of an isolated and dedicated network to each service provider, infrastructure providers must employ appropriate scheduling algorithms in all of the network elements.
Existing system virtualization technologies provide efficient scheduling mechanisms for CPU, memory, disk, and network interface in each of the virtual machines running on the host machine [105]. Network virtualization can ex- tend these mechanisms to implement resource scheduling in the physical infrastructure. Previous results from re- search on packet scheduling algorithms for IP networks can also be useful in the design of schedulers.
5.8. Naming and addressing
Due to potential heterogeneity of naming and addressing schemes in coexisting virtual networks, end-to-end communication and universal connectivity is a major challenge in a network virtualization environment. In addition, end users can simultaneously connect to multiple virtual networks through multiple infrastructure providers using heterogeneous technologies to access different services, which is known as über-homing [81]. Incorporating sup- port for such heterogeneity in multiple dimensions is a fundamental problem in the context of network virtualization.
Recently proposed iMark [81] separates identities of end hosts from their physical and logical locations to add an additional level of indirection and, with the help of a global identifier space, provides universal connectivity without revoking the autonomy of concerned physical and virtual networks. However, while conceptually possible, iMark is not physically implementable due to exces- sive memory requirements. Therefore, one key research direction in naming and addressing is to find a viable global connectivity enabling framework.
5.9. Dynamism and mobility management
Network virtualization environment is highly dynamic. At macro level, virtual networks with shared interests can be dynamically aggregated together to create federation of virtual networks. Multiple federations and virtual net- works can also come together to form virtual network hierarchies[81]. Aggregation and dissolution of control and data planes (e.g., naming, addressing, routing, and for- warding information) for macro level dynamism is an unresolved issue.
At micro level, mobility of end users from one physical location to another and migration of virtual routers for operation and management purposes [75] poses the biggest challenge. Finding the exact location of any resource or end user at a particular moment and routing packets accordingly is a complex research challenge that needs efficient solution. In addition, network virtualization al- lows end users to move logically from one virtual network to another, which further complicates the problem.
5.10. Virtual network operations and management
Network operations and management has always been a great challenge for the network operators. Division of accountability and responsibilities among different participators in a network virtualization environment promises increased manageability and reduced scopes for error [3]. Keller et al. [76] propose proactive and reactive mecha- nisms to enforce accountability for hosted virtual networks.
Considerable flexibility must be introduced from the level of network operations centers (NOCs) to intelligent agents at network elements, to enable individual service providers configure, monitor, and control their virtual net- works irrespective of others. The concept of MIBlets [60] used in VNRMS to gather and process performance statis- tics for each of the coexisting virtual networks instead of using a common MIB can be a good starting point.
Since a virtual network can span over multiple underlying physical networks, applications must also be developed to aggregate information from diverse, often conflicting, management paradigms followed by participating infra- structure providers. Introducing a common abstraction layer, to be followed by all the management softwares, can be an effective solution [106].
Failures in the underlying physical network components can give rise to cascading failures in the virtual net- works directly hosted on those components. For instance, a physical link failure will result in failures of all the virtual links that pass through it. Similarly, any physical node failure might require re-installations of all the service provider’s custom softwares. Detection and effective isolation of such failures as well as prevention and recuperation from them to stable states are all open research challenges.
5.11. Security and privacy
Even though network virtualization strives for isolation of faults and attack impacts, it does not necessarily obviate existing threats, intrusions, and attacks to physical and virtual networks. In fact, to some extent, network virtualization gives rise to a new array of security vulnerabilities. For instance, a denial-of-service (DoS) or a distributed DoS (DDoS) attack against the physical network in a virtualized environment will affect all the virtual networks hosted on that network. Programmability of network elements – powerful and expressive in trusted hands – can in- crease vulnerability if there are security holes in programming models. To avoid such pitfalls, recent proposals (e.g., CABO) argue for controlled programmability by trading off flexibility for security without any definitive answer to permissible levels access to programmable hardware.
A detailed study of possible security vulnerabilities can give insights into developing programming paradigms [107]and virtualization environments that are secure and robust against known attacks. Established secured tunneling and encryption mechanisms (e.g., IPSec [15]) in VPNs can also be used in this context to increase security and enforce privacy.
Being translated, please wait..
