Multiple SDN applications can make several harmful interferences unintentionally, although each individual application may be properly developed. This paper proposes a Multiple SDN Applications Interference Detector (MSAID). To bridge the gap between the source code of applications and the actual interferences, we leverage symbolic execution and constraint solving to obtain how the event handler handles the input messages. We then analyze the complex interaction of multiple applications and present novel methods to identify the interferences. Finally, we evaluate its correctness and prove its usefulness with a series of SDN applications.

The programmability of Software-Defined Networking (SDN) challenges the correctness and reliability of networks. There may be design flaws as well as implementation bugs in SDN applications. White-box testing methods with formal models rely on source codes, which limits the applicability of these methods. Black-box methods without behavior models cannot systematically cover an application's functions. Most previous work has mainly focused on design flaws and has ignored implementation bugs. In this paper, we propose a new black-box test framework to detect both design flaws and implementation bugs. Following this test framework, we propose a new model, Information Table Extended State Machine (IT-EFSM), combining a group of parallel state machines and an abstract topology to specify the SDN applications. We employ a model checking tool to generate tests against design flaws and propose a test generation based on partial composition, symmetry simplification on the topology and topology simulated execution to expose implementation bugs. The experimental results of the testing process demonstrate the effectiveness and applicability of our method.

Existing tools for Software-Defined Networking (SDN) data plane testing can be classified into two classes: white box and black-box. For the former, all or part of source codes should be accessed. But for testers outside the manufacturers, the accessing of source code is impossible or very difficult in most cases, especially for hardware devices. For the latter, test cases are manually developed, which cannot ensure the coverage. In this paper, we present a model based black-box systematic testing method for SDN data plane. We propose a new model, Pipelined Extended Finite State Machine (Pi-EFSM), to specify the multiple-level pipeline of SDN data plane. For the Pi-EFSM model, we present a 3-phase systematic test generation approach. By using a hierarchical test generation strategy, the proposed test generation method can alleviate state space explosion to some extent. Our test generation method can achieve the systematic coverage towards the elements of the model. We apply our method in the testing of Open Flow switches (specification version 1.3.0). We build a Pi-EFSM model for Open Flow switches and derive the executable test sequences. Some implementation faults and specification confusions are exposed when we test two switches, Open switch 2.1.0 and CPqD Open Flow 1.3 Software Switch.

Current intra-domain routing protocols computes only shortest paths for any pair of nodes which cannot provide good fast reroute when network failures occur. Multipath routing can be fundamentally more efficient than the currently used single path routing protocols. It can significantly reduce congestion in network by shifting traffic to unused network resources. This improves network utilization and provides load balancing. To enhance failure resiliency we propose a new scheme More Nodes Have At Least Two Choices (MNTC) where the goal is how to maximize the number of nodes that have at least two next-hops towards their destinations. We evaluate the algorithm in a wide space of relevant topologies and the results show that it can achieve good reliability while keeping low stretch.

The Internet is playing an increasingly crucial role in both personal and business activities. Handling link failures is an important task in designing routing protocols. To enhance the network availability without incurring significant extra overhead, we propose a novel link protection algorithm, Hybrid Link Protection Scheme (HLP) to achieve failure resilient routing.

As the soul of the Software-Defined Networking (SDN), the quality of control plane applications determines the reliability of the networks. Unfortunately, better programmability in SDN increases the risk of bugs and challenges for testing. Because manually testing seems to be inefficient, automatic testing methods become promising alternative. Both white-box method with models and black-box method without model have limitations. In this paper, we propose a formal model for blackbox testing of SDN applications. We use a group of components to describe the data structure stored in the applications and the system behaviors. It is easier and more natural to specify applications. Based on our models, we present our work-in-progress testing framework. It can iteratively improve the design model with model verification and expose implement bugs with model-based testing.

Current researches on model-based testing mainly focus on the single component model, such as FSM (Finite State Machine) and EFSM (Extended FSM). To model and test parallelism and concurrency among different protocol components, traditional CFSM (Communicating FSMs) models communication as asynchronous message exchange, which is not suitable for the scenario that parallel protocol components read shared variables from each other. We have developed Parallel Parameterized Extended Finite State Machine (PaP-EFSM) to handle this situation. In this paper, we present a hierarchical test generation approach for PaP-EFSMs based on reachability graphs. The combination of bottom-up reachability graph generation and top-down test sequence generation ensures the executability of the test sequences and alleviates state explosion. We apply this method to the conformance testing of SAVI, a real protocol for anti-spoofing of IP source addresses. We build a set of PaP-EFSMs for SAVI and derive the executable test sequences, which expose many implementation faults when applied to real devices from 4 different vendors.