ENDEAVOUR @ Looking Beyond the Internet
A workshop on Software-defined Infrastructure and Software-defined Exchanges took place in Washington D.C. on 4-5 Feb 2016 as part of the NSF “Looking Beyond the Internet” series of workshops. The purpose of this workshop was to recommend a 5-year agenda for research programs in new technology paradigms that will radically transform the Internet, and to identify the needs for infrastructure to support that research.
Marco Canini gave the second keynote address. The talk is summarized below in this article and the slides are posted here.
“SDN is dead. Long live SDX!”
Marco Canini, Université catholique de Louvain
The talk started with a retrospective on the status of SDN and the recent research projects on SDXes that laid the starting ground on architectures, capabilities and use cases of SDXes. Despite general interest by network operators, there is evidence that SDXes must mature through much additional research before their disruptive potential can be fully realized. Hence, the second part of the talk focused on what major research challenges exist in the 5-10 years agenda, urging the audience to think about the need for fundamental research that deals with formal foundations about SDXes as well as equally important applied, interdisciplinary research that investigates the role of SDXes in addressing major societal problems.
The talk started by arguing that with SDN the networking research and operators communities have scraped the tip of a large iceberg and it's now time to see what's beneath the water. SDN has reshaped many networks, but in particular most success stories have happened in the datacenter, where it is easier to innovate and deploy.
Motivated by the desire to bring this innovation to wide-area traffic delivery, a natural place for this to occur is at IXPs, which are becoming increasingly prevalent; thus, maximizing the effect onto inter-domain routing of even a single SDN deployment.
Recent projects on Software-Defined Exchanges (SDXes) such as Google's Cardigan, SDX @ Princeton, and ENDEAVOUR have shown feasibility of early ideas in this space. They established ways to incrementally deploy SDN at IXPs, offer more explicit control over inter-domain routing, and address various scalability challenges that exist in large IXPs with hundreds of participants. All of this while maintaining good interoperability with BGP.
The talk argued that the work done so far is necessary "plumbing" and that we need to see more deployments and testbeds in this area. Though future work should stay conscious that technical innovations have to play well within the overall incentives and business aspects of the inter-domain settings, which are inherently more challenging than single-domain settings like datacenters. In support of this argument, the talk briefly presented some anecdotal evidence of the mismatches between opportunities enabled by fine-grained, programmatic traffic control versus current best practices.
The second section of the talk focused on the bold vision set forward by the workshop towards the concept of SDXes that will enable large-scale interconnection of SDIs. The talk discussed a rich research agenda on the road to this vision, highlighting the following:
- Security: Increased Internet security is a goal highly sought after and a desired use case by operators, in particular, detecting and preventing DDoS. Beyond DDoS mitigation, SDXes might enable new architectures that can help by design address network attacks. For example, they provide an opportunity to reconsider the line of research on network capabilities as they might be a way to embed costs into traffic, such that costs can act as a deterrent for attackers.
- Privacy: Beyond improving inter-domain routing, there is a clear expectation of deploying network functions at SDXes such as caches, optimizers, packet scrubbers, etc. This raises questions regarding the privacy of processing traffic at exchanges and the neutrality of SDXes. Who controls the network functions? Who specifies what traffic traverses which network functions? Is there any auditing that we can perform?
- Business Confidentiality: Generally, policies used by ISPs for their peering agreements and route selection are private. SDXes need to provide rich services that can be consumed by participants through APIs without this causing leakage of any confidential information to other participants at the SDX.
- Reliability & Robustness: SDXes will be critical infrastructure that needs to be dependable. Yet this is made challenging by the growing presence of software in networks. Today many experts from the networking and formal methods communities believe that despite their importance, tools for programming and reasoning about networks are still in a state of infancy.
- Quality-of-Experience (QoE): SDXes are well suited to fill a role in increasing QoE in the Internet. Existing problems and inefficiencies largely stem from the lack of information exchange between application providers and ISPs, which can be detrimental for the user. SDXes might enable a bi-directional flow of information between these parties and act as a trusted optimizing arbiter in the presence of contrasting objectives.
- Marketplace: SDXes prove the unprecedented opportunity to enable an efficient marketplace at the level of transport and routing. There are several forms this can be envisioned: (1) a marketplace for setting up peering on-demand through fast commercial negotiations and connectivity setup, (2) a marketplace for optimizing routing and providing end-to-end guarantees, leveraging the improved network visibility that SDXes have, (3) a marketplace for 3rd party providers of virtualized network functions (e.g., security providers) to tap into the ecosystem.
The talk closed by highlighting the globally disruptive potential for innovation with SDXes: How do we turn SDX into a platform upon which people can build solutions to solve world's problems? Major ongoing trends such as smart grid, smart transportation, smart cities, electronic voting, green economy, online education require better infrastructure as the Internet is ill suited for many new requirements of these technological innovations. The talk posited that SDXes fit into these themes by giving three examples.
- SDXes for resource fluidity: Connecting hundreds of networks each, an interconnection of SDXes can become a key enabler for better sharing of bandwidth and other resources.
- Crowd-sourced SDXes: a global network of SDXes can ensure that independent networks can be quickly constructed and operated when times require it (e.g., natural disasters).
- SDX vaults: in the IoT era with massive data acquisition capabilities, SDXes stand to act as a neutral common ground providing brokering services for privacy-preserving data aggregation and analysis between producers and consumers of such data.
Meeting these goals will require both fundamental research dealing with formal foundations for reasoning about SDXes as well as applied, interdisciplinary work that blends together system prototyping, deployment, testbed operation with use cases and long term vision about the potential impact of SDXes.