March 24 – 27, 2013, Dagstuhl Seminar 13131
1 / 2 >
For support, please contact
The recent vision of the "Future Internet" attracts significant networking research and causes controversial debates on the actions to be taken. Clean-slate initiatives envision a fresh start that put fundamental principles of networking into question. Avoiding any constraints of the current Internet implementation, the ambition of the clean-slate approach is to understand and design the `right' network architecture. Evolutionary approaches, on the other hand, seek incremental improvements, assuming that the Internet can --as in the past-- be fixed to accommodate the changing needs of users and applications.
Numerous initiatives on the Future Internet, like FIND, GENI funded by the NSF, FIRE, 4WARD by the EU, and G-LAB by the BMBF, reflect the importance of the topic. Characteristic for numerous Future Internet initiatives is an experimental approach using testbed facilities such as the GENI or the G-Lab platform.
Challenges that are of central importance for the Future Internet fall into the following categories:
- Network design: computer networks and the Internet obey certain architectural guidelines that reflect experience gained in the art of network design, such as layered reference models or the Internet end-to-end argument. While these principles are backed up by the success of the Internet, it has to be noted that the network exhibits major architectural restrictions, e.g., regarding mobility, security, and quality of service. Computer networking as a relatively young field of research can benefit significantly from architectural reconsiderations that are initiated by clean-slate initiatives. While today, network theory is largely descriptive, this Dagstuhl seminar investigated the potentialities of a prescriptive network theory, which could justify a methodical rule/equation-based approach for the design of future networks.
- Virtualization: the virtualization paradigm revolutionized the use of computers and data centers, where the flexibility and mobility of virtual machines offers tremendous potential, posing, however, significant new challenges for networking. On the other hand, the virtualization paradigm has already many applications in networking, e.g., in virtual private networks or overlay networks. Currently, virtualization finds its way into network components, e.g., routers, itself, where the decoupling of logical entities from the physical substrate enables major innovations, e.g., concurrent (possibly post-IP) networks, infrastructure as a service, redundant shadow configurations, in network management, and in energy efficiency. Furthermore, the provisioning of service-oriented virtual networks across multiple infrastructure providers creates the need for separation between the network operations and the physical infrastructure. This is expected to change the way that virtual networks are managed, debugged, and operated. The Dagstuhl seminar contrasted different approaches to network virtualization and investigated their applications.
- Experimental research: the Internet standardization process relies on running code and real world verification. An essential prerequisite for the transfer of research results is building of large scale testbed networks. These are frequently implemented as virtual, Software Defined Networks that run concurrently to a production network using the same hardware. The Dagstuhl seminar revisited the experimental approach and gathered lessons learned and best practices.
During the seminar, we discussed and (partly) answered the following questions:
- Is a prescriptive network theory feasible?
Today, network research is largely descriptive, e.g., there exist methods and tools to model communication networks and protocols, to analyze their performance, or to verify their correctness. The design of new networks, however, lacks a prescriptive network theory that provides necessary rules and equations that specify how a network for a given purpose has to be built. Instead, network design relies heavily on previous experience and best practices frequently resulting in incremental works. In contrast, the clean-slate Future Internet approach seeks to build a new Internet architecture from scratch. In this case the design space is entirely open requiring decisions regarding functional and non-functional aspects, e.g.,
- Where to implement reliable/unreliable and connectionless/connection-oriented?
- Where (end systems or network) and in which layer to keep state information?
- Where and how to achieve security, quality of service, and dependability?
- How to split locators and identifiers?
Given the examples above, we discussed:
- How can a prescriptive approach to network theory be formulated?
- What are the perspectives and the fundamental limits of the candidate approaches?
- What are the prospects of the approach if successful?
- Which insights can the experimental, testbed-based approach reveal?
Many approaches to the Future Internet are experimentally driven and centered around a testbed that ideally if successful becomes the first running prototype of the Future Internet. Clearly, testbeds are indispensable to implement running code as a proof-of-concept, whereas their use for understanding networking and for establishing new principles and paradigms can be debated. In the seminar we elaborated on this question to provide answers to:
- Which insights can be expected?
- Which exemplary fundamental insights did emerge from testbeds?
- For which use cases are testbeds meaningful, e.g., to engineer details, to approach concepts weakly understood, to understand the impact of users, etc.?
- How should a testbed platform look like, which properties must be provided?
- How can testbeds be benchmarked to achieve comparability and validity?
- What are the challenges for wide-area service-oriented virtual networks? The virtualization paradigm gained a lot of attention in networking as it provides numerous useful applications and promises to solve a number of important issues, such as the gradual deployment of new networking solutions in parallel to a running production network. Considering existing networking technologies, it becomes apparent that virtual networks and virtual network components are already being used in a multitude of different ways and in different layers, e.g., Virtual LANs (VLANs), Virtual Private Networks (VPNs), the Virtual Router Redundancy Protocol (VRRP), or in form of overlay networks to name a few. Furthermore, the abundance of resources offered by commodity hardware can turn it into a powerful and highly programmable platform for packet processing and forwarding. The virtualization of such programmable network elements can provide network slices which are highly customized for particular network services and applications. The topics that were discussed at the seminar include:
- Resource discovery and provisioning of virtual networks across multiple domains, given that infrastructure providers will not be willing to expose their topology, resource information and peering relationships to third-parties;
- Virtualization of network components (e.g., resource allocation, isolation issues);
- Scaling of virtual resources to meet variable service demand;
- Use cases of network virtualization.
This report provides an overview of the talks that were given during the seminar. Also, the seminar comprised a one minute madness session for introduction and for statements on the Future Internet, a breakout session for group work on the topic of prescriptive network theory, as well as podium discussions on experimentally driven research and on the use cases of SDN. The discussions, viewpoints, and results that were obtained are also summarized in the sequel.
We would like to thank all presenters, scribes, and participants for their contributions and lively discussions. Particular thanks go to the team of Schloss Dagstuhl for their excellent organization and support. We also would like to thank Anil Madhavapeddy for his feedback and comments on SDN.
- World Wide Web / Internet
- Future Internet
- Clean-slate Design
- Virtual Networks