This project is aimed at the design and experimental validation of a comprehensive clean-slate future Internet architecture. The proposed MobilityFirst architecture is motivated by the ongoing paradigm shift of Internet usage from today?s fixed PC/host (client)?server model to emerging mobile data services and pervasive computing applications. The major design goals of the architecture are: mobility as the norm with dynamic host and network mobility at scale; robustness with respect to intrinsic properties of the wireless medium; trustworthiness in the form of enhanced security and privacy; usability features such as support for context-aware services, evolvability, manageability and economic viability. The key components of the MobilityFirst network design are: (1) separation of naming and addressing, implemented via a fast global dynamic name resolution service; (2) self-certifying public key network addresses to support strong authentication and security; (3) generalized delay-tolerant routing with in-network storage for packets in transit; (4) flat-label internetwork routing with public key addresses; (5) hop-by-hop transport protocols operating over segments rather than an end-to-end path; (6) a separate network management plane that provides enhanced visibility; (7) optional privacy features for user and location data; and (8) an integrated computing and storage layer to support programmability. The project?s scope includes architectural design, validation of key protocol components, testbed prototyping of the MobilityFirst architecture as a whole, and real-world protocol deployment on the GENI experimental infrastructure. The results of this project will provide architectural guidance for cellular-Internet convergence, and are expected to influence future technical standards in the networking industry.
Our main outcome in this project is the design, implementation, and integrationof the compute layer in the MobilityFirst architecture. We designed the compute layer to distributed ``cloudlet'' computeclusters co-located with an ISP's PoPs (Point of Presence). Each ISPuses a conceptually centralized control to control all computeresources in its network. An ISP offers various network functions toend customers or application-level service providers using its computelayer. Sample services include, but not limited to: content caching,unwanted traffic filtering, anonymous communication, load balancing,and traffic optimization. We implemented the computer layer and integrated it into the overallMobilityFirst architecture. We evaluated the performance of ourimplementation both in a testbed environment and on the Internet. Ourresults show that the computer layer shortens the latency of severalnetwork applications such as rate adaptive video delivery. Specially, our work includes the following: - We published the preliminary design of the compute layer (which we call PacketCloud) in the in the MobiArch'13 workshop. We are currently submitting the design, implementation, and evaluation of the compute layer to a journal. - We worked together with the Rutgers team, and integrated our PacketCloud part into the mainstream MobilityFirst code base. - With the compute layer, we are able to add various in-network services to the MobilityFirst-based network environments. As a demonstrative example, we have built a rate-adaptive video delivery system the GENI testbed, and demonstrated the implementation at the 20th GENI Engineering Conference (GEC-20). - We conducted a measurement-based study using the PlanetLab and the Amazon EC2 datacenters. Our study shows that using the compute layer to provide in-network services has significantly lower latency than using public datacenters. - We implemented a number of PacketCloud use cases, including content cache, a firewall, anonymous communication modules, a VPN gateway, a load balancer, and a traffic optimizer. - We conducted a detailed evaluation for the per-hop latency and scalability of the compute layer. Our study shows that PacketCloud can handle high throughput data traffic, and adds a negligible per-hop latency. - We deployed PacketCloud on the real Internet using the PlanetLab testbed. We used a global content delivery example to demonstrate the benefit of the compute layer.
