No abstract

Unreliable failure detectors are a basic building block of reliable distributed systems. Failure detectors are used to monitor processes of any application and provide process state information. This work presents an Internet Failure Detector Service (IFDS) for processes running in the Internet on multiple autonomous systems. The failure detection service is adaptive, and can be easily integrated into applications that require configurable QoS guarantees. The service is based on monitors which are capable of providing global process state information through a SNMP MIB. Monitors at different networks communicate across the Internet using Web Services. The system was implemented and evaluated for monitored processes running both on single LAN and on PlanetLab. Experimental results are presented, showing the performance of the detector, in particular the advantages of using the self-tuning strategies to address the requirements of multiple concurrent applications running on a dynamic environment.

Software-defined networks (SDN) usually rely on a centralized controller, which has limited availability and scalability by definition. Although a solution is to employ a distributed control plane, the main issue with this approach is how to maintain the consistency among multiple controllers. Consistency should be achieved with as low impact on network performance as possible and should be transparent for controllers, without requiring any change of the SDN protocols. In this work, we propose VNF-Consensus, a virtual network function that implements Paxos to ensure strong consistency among controllers of a distributed control plane. In our solution, controllers can perform their control plane activities without having to execute the expensive tasks required to keep consistency. Experimental results are presented showing the cost and benefits of the proposed solution, in particular in terms of low controller overhead. | INTRODUCTIONSoftware-defined networks (SDN) separate the control plane from the data plane, which improves their flexibility, programmability, and management. 1,2 The control plane is usually centralized, consisting of a single controller, while the data plane consists of numerous network devices distributed across the network. While a centralized approach is attractive as it is simpler to operate and manage, it represents a vulnerability as the controller is a single point of failure with a direct impact not only in terms of resilience (i.e., if the controller fails the whole network stops running) but also on performance and scalability 3 (i.e., the single controller has to process all requests from all switches). The solution is to distribute the control plane, employing multiple controllers that share responsibilities. 4 Several different distributed SDN control plane strategies have been proposed. 5,6 Although the advantages of employing multiple controllers should be clear, there is also a cost to pay: in order to employ a distributed control plane it is necessary to employ techniques to guarantee the consistency among the