Intelligent Automation, Inc. — Department of Energy SBIR Phase I: Designing an efficient and scalable network fault diagnose tool for such a complex system

Description

Designing an efficient and scalable network fault diagnose tool for such a complex system is facing the following challenges: 1) Inherently unpredictable and fault-prone status in the involved networks for scientific collaboration and science mission due to the dynamic nature of network and service demands, and the increasing complexity of configuration options resulting from the implementation of Quality of Service, routing, and information assurance. 2) The DOE networks are typically large, geographically distributed, and constantly evolving with various applications and services which require the support of complex hardware and software components. Science mission requirements, applications/services and the underlying network capabilities are inter-connected. Consequently, the system will be associated with a large number of complex failure modes involving interactions across various system components in different levels. Even a small problem in the network along with any part of an end-to-end path can potentially degrade the user experience significantly, and those issues are hard to detect and time consuming to isolate and fix. 3) Integrated domain knowledge on network and inter-layer interactions is needed. Many network fault diagnose tools or algorithms were designed for the traditional wired networks, or hard faults of the devices, faults and symptoms at the application level, or focusing on detecting a specific fault. They either cannot be directly applied to the DOE networks, or cannot provide the full support for the DOE networks. Intelligent Automation, Inc. (IAI), along with Professor Nasir Ghani from the University of South Florida, and ESnet Advanced Networking Technologies Group at Lawrence Berkeley National Laboratory, proposes to develop NetFaultSONAR a scalable and effective network fault detection and localization scheme. The proposed NetFaultSONAR can analyze the monitored network as a whole and assist users (i.e., network operators) in maintaining, optimizing, and troubleshooting the network. It can take advantage of existing perfSONAR framework, by taking collected data for existing tools/services as the input for our fault detection and localization components. The goal is to provide a tool that can monitor important network performance in real-time or near-real-time, automatically collect/retrieve the network data/status/configuration, perform cross-layer multi-domain fault detection and analysis to pinpoint the root cause, and assist the user (i.e., network operator) to determine the most efficient action to fix the issues accordingly. It will significantly improve the reliable access to DOE networks, and reduce the cost and risks for network management. Commercial Applications and Other Benefits: The proposed network analysis tool has the potential to greatly reduce overall operational costs, whilst maintaining or even enhancing the reliability of existing DOE networks. Moreover, due to the heterogeneous and complex nature of scientific networks, the proposed network fault analysis solution can be applied to the operation of a full range of DOE and partner networks, i.e., including ESnet local networks at DOE national laboratories, and those at numerous collaborating institutions and universities across the world. In addition to DOE and partner networks, a broad range of companies and organizations can also benefit from our proposed tools, such as next generation Internet backbone network, DOD military networks, etc.