Date: March 1, 2012
Researchers are exploring a fundamentally new approach to networking. Dubbed data-centric networking, information-centric networking, or name-oriented networking, it could transform the Internet infrastructure if it ends up improving the performance and efficiency of content delivery networks (CDNs.)
The Internet Society held a panel discussion concurrent with the IETF meeting in Taipei, Taiwan, in November to discuss the problems that are driving interest in data-centric networking and the implications of this research if it is deployed.
Leslie Daigle, chief Internet technology officer at the Internet Society, served as moderator for the high-level discussion entitled “Beyond PUT, POST, and GET: Application data routing carves its own path.’’
Daigle pointed out that CDNs, which are typically proprietary, have been involved in some IETF standards work, particularly in the Content Delivery Networks Interconnection working group.
According to Daigle, the new data-centric networking approaches, however, could have “many and varied” implications for the Internet infrastructure. “Unlike today’s Internet paradigm, they do recognize data storage as a first-class entity, and they are typically independent of the topology of the network that we understand,” she said.
Network operators say that dealing with large volumes of content, particularly streaming video, in a cost-effective manner is one of the toughest challenges they face today.
“We have a lot of data crossing my own company’s network,” said Rich Woundy, vice president of platform architecture at Comcast Cable. “We want to make sure that we use the network as efficiently as possible and also provide a good customer experience.”
Woundy explained that the CDN approach of caching content and distributing it across the Internet has created a good end-user experience and has been cost-effective for content providers. But the uptick in video traffic—which consumes about 60 percent of all bandwidth in the United States—is making the current content distribution approach too expensive for network operators. In addition, the way that CDNs handle failover for busy servers can create inefficient traffic routing.
Content distribution “is consuming a lot of our network. It’s getting our attention, and it’s getting our CFO’s attention,” Woundy said. “The question is: What can I do to balance my need to grow my network while fulfilling the needs of content providers, and making it work efficiently with the other CDN providers out there? … In many cases, the optimisation at the CDN layer and the optimisation at the core and backbone layer is not necessarily in sync.”
Aaron Falk, director of business and product management for CDN Solutions at Verivue, agreed that network operators are feeling the pinch of providing network bandwidth for video streaming applications. “Neither Internet service providers (ISPs) nor mobile operators have the upstream capacity to meet the bandwidth demand created by popular video applications,” he said.
“If everybody is at home watching Netflix on a Friday night, this creates a real burden on the operators, and it’s a burden that they get no additional revenue for.” Falk says. “They see their costs go up, and their subscription fees stay flat. So this is a problem they are interested in seeing solved.”
Falk says that the Internet building blocks available to CDNs and network operators to solve this problem—Uniform Resource Identifiers (URIs), Border Gateway Protocol (BGP), and ping times—are coarse rather than fine-grained tools for connecting a client to a nearby content source.
“Because that’s not what these architectural elements were designed to do, there are all sorts of inefficiencies and problems that arise,” Falk says. “It’s basically trying to build a system to adjust performance issues with mostly found pieces of technology … That motivates a bunch of research.”
Dave Oran, a Cisco Systems fellow, said CDNs are far down the path of optimising systems to deal with large amounts of content that the Internet architecture wasn’t designed to carry and so handles in a suboptimal way. That’s why there is promise in researching new architectures, such as data-centric networking, to solve this problem.
“The notion here is that you directly name content … and not worry about what hosts they are coming from and where they are going,” Oran says. “You base routing on data … and security on data, and you base optimisation of the network on placing storage at strategic places in the network so data can visit and live on that server.”
One requirement for data-centric networking is self-certifying data, in which data items have signatures to assure that they are in the exact format created by the publisher of the data.
Oran identified two primary camps of data-centric networking research: the first is what he calls the dessert-topping approach and the second he refers to as the floor wax-approach. With the dessert-topping approach, researchers are trying to keep the current Internet architecture—including hosts, IP addresses, and HTTP—and build an extra layer on top of this infrastructure to handle data naming, storage, and security. With the floor-wax approach, researchers are considering replacing the entire Internet infrastructure (except IP) with a newly created system that is optimised for data-centric networking.
The floor-wax approach eliminates many middleboxes including DNS (domain name system) load balancers, URI (uniform resource identifiers) redirectors, cooperative cache engines, and transparent proxy caches
“The floor-wax approach says let’s go back and reengineer all the way back to IP,” Oran explained. “This allows us to look at the fundamentals of naming, the fundamentals of routing, the fundamentals of security, and the fundamentals of robustness and scalability.”
Oran said large-scale systems using data-centric networking will be built during the next two or three years. “This is an area worth watching because it’s highly disruptive and has the opportunity to dramatically simplify Internet architecture,” he added.
Falk pointed out that in order for data-centric networking to work it needs to not only tackle technical problems of video distribution but also support the complex business relationships between content owners, content distributors, network operators, and end users. “We should find a way to build technical solutions that allow these business relationships to work,” he said.
Oran added that it’s important for researchers who want to fundamentally change the Internet infrastructure to support data-centric networking to understand that they shouldn’t optimise for one application—video—when other applications, such as very large distributed databases, may become important in the future.
“We are looking at these new named data architectures for a broad range of applications … not just the pain point of today,” Oran said.