CRII: NeTS: Towards a database-defined network

Temple University, Philadelphia PA

Investigators

Abstract

Software-defined networking (SDN) provides an unprecedented opportunity to exercise computing principles in simplifying networking practice. Indeed, many salient SDN features --- central control and programming interfaces --- are enabled by principles from the fields of programming language, operating systems, and distributed systems. While these principles have produced a variety of programming abstractions that drastically improve networking practice, reducing the complex distributed protocols to simpler centralized programs, there is little consensus on what are the right abstractions. Moreover, network architects today need to combine heterogeneous data sources --- network forwarding rules, data flow and QoS metrics, host intrusion alerts, and so on --- to produce a cohesive view of the network and investigate problems. To tackle the challenging network abstraction and integration problem, this project will explore the computing principle --- a principle blessed with decades of fruitful theories and commercial success, yet surprisingly under-investigated in SDN --- of database. This database-defined network project builds on the perspective that SDN control fundamentally revolves around data representation, where the entire SDN network is under the hood of a standard SQL database, relying on SQL for flexible and extensible network abstraction, and leveraging the database integration techniques for combining multiple control applications into a coherent forwarding behavior. In achieving this ambitious goal, this project will first develop a novel database-centered controller platform that address two unique networking challenges not found in classic database applications: (1) the system extensibility requirement to accommodate an enlarging body of control applications; and (2) the performance requirement to cope with very fast data-plane updates under complex control applications. In addition, the controller platform will be enhanced with data integration component to manage dependencies between control applications, addressing the long-standing problem of integrating control applications --- combining a collection of potentially overlapping and conflicting modules into a coherent network forwarding behavior. The results of the project will yield solutions to multiple problems, including an attractive form of evolvable SDN abstractions and automated coordination between controls applications. All the abstractions and coordination facilities will be exposed to application programmers, network engineers and architects via the broadly familiar SQL interfaces. Thus, the project is likely to be valuable both technically and as a way to encourage more rapid uptake of SDN. The project will integrate the theoretically-grounded research methodology and content of this proposal with education by developing a graduate-level 'Computing principles in software-defined networking' course. Informed by the analysis and comparison of the various fundamental approaches (e.g., database, distributed system, programming language) to prominent SDN problems of abstraction and integration, the project will also result in a survey paper on computing principles and SDN.