This project presents a new paradigm-shift approach in fault diagnosis by investigating network problems without requiring any monitoring sensors or active measurements, and assuming little or no knowledge about the network. The goal is to develop accurate, scalable and cost-effective network problem diagnosis that reason about uncertainty in case of incomplete knowledge without intrusive active probing or network monitoring. This project investigates a novel approach that uses evidential reasoning based on user observations to analyze the end-user views as evidence and compute a combined belief for determining the most possible root causes in overlay networks at real-time. The project also investigates techniques to rank the overlay paths based on their quality. The reasoning results can then be fedback into adaptive active monitoring, and dynamic virtual assignment/reconfiguration systems to optimize problem monitoring and recovery, respectively.
Developing techniques and tools that enable sharing and analyzing end-host observations provide powerful diagnosing capabilities to service providers, system developers, and administrators to in problem determination, characterizing network conditions, configuration debugging and troubleshooting. These techniques are applicable on both overlay and traditional networks. This project enables trained workforce in this area through teaching and supervising students.
The main goal of this project is to develop accurate, scalable and cost-effective overlay network (a.k.a. Virtual Network (VN)) problem diagnosis that can reason about uncertainty in case of incomplete knowledge without intrusive active probing or monitoring sensors. Our more specific goals were: - Develop network embedding schemes that allow for efficient network fault diagnosis - Develop network diagnosis schemes based on end-to-end passive monitoring - Develop techniques for virtual network migration to help in mitigating poor network performance uncovered through diagnosis or to improve the diagnosability of VNs. We believe this work has significant impact because network virtualization is likely to become a primary interface on the global network infrastructure seen by applications and users. Our work will help enable this vision by facilitating the maintenance and operation of such an infrastructure. Further, we envision the need of 'Virtual Network Providers' as commercial entities. Such providers will need a repertoire of mechanisms and tools for efficient placement, diagnosis and migration as developed in our work. Our work consisted of the following activities: 1- Overlay Placement for Diagnosability Overlays must be monitored for various kinds of problems so that efficient performance can be sustained. An overlay’s topology and placement on the substrate have a considerable effect on the level of difficulty in monitoring it. We study the problem of placing overlay networks onto the substrate in a way that makes it easier to detect and localize faults, in other words, improves their diagnosability. 2- Fine-Grain Diagnosis of Overlay Performance Anomalies Using End-Point Network Experiences We develop a novel diagnosis technique to localize performance anomalies and determine the packet loss contribution for each network component 3- Design and Analysis of Schedules for Virtual Network Migration We found that overlay placement can affect its diagnosability. The question remains of how to achieve a desired placement. This led us to the question of what are the mechanisms that can be used to adjust the placement of an overlay. We considered the problem of the design of a mechanism for virtual network movement. 4- Virtual Network Migration for PlanetLab While VN placement and migration has been studied extensively in the past, not much attention has been devoted to deploying and evaluating these functions over a real infrastructure. In this activity, we study the VN migration problem based on network virtualization in PlanetLab. We create a tool, PL-VNM, which orchestrates the VN migration on PlanetLab for a given new VN placement. 5- Virtual Network Placement in Software-Defined Networks (SDN) Software-defined networking has emerged as a powerful approach to improve the customizability and flexibility of networks. Our goal was to consider questions of VN fault diagnosis and fault mitigation through VN migration. However, there was no VN embedding techniques for SDN networks studied in the literature. As a result we decided in this work, to focus on virtualization in the realm of software-defined networks (SDNs), and study how to embed virtual networks in this environment. In SDNs, the presence of a central controller is a complicating factor, and customizable routing and differences in resource sharing present new opportunities and challenges. Educational Impact: Two PhD students worked on this project, one male and one female. One of the students completed his degree requirement in August 2013. The Project PI teaches a unit on overlay/virtual networks in the first level graduate course (CS6250). Results from this work and earlier work in this area supported by NSF are included in the class discussion. Software: A prototype network migration controller for Planet Lab (PL-VNM) will incorporate our mechanisms and heuristics for network migration. The tool is in the final stages of development and we will make this tool widely available once it is completed.
