Towards Automated Network Configuration Management

Modern networks are designed to satisfy a wide variety of competing goals related to network operation requirements such as reachability, security, performance, reliability and availability. These high level goals are realized through a complex chain of low level configuration commands performed on network devices. As networks become larger, more complex and more heterogeneous, human errors become the most significant threat to network operation and the main cause of network outage. In addition, the gap between high-level requirements and low-level configuration data is continuously increasing and difficult to close. Although many solutions have been introduced to reduce the complexity of configuration management, network changes, in most cases, are still manually performed via low--level command line interfaces (CLIs). The Internet Engineering Task Force (IETF) has introduced NETwork CONFiguration (NETCONF) protocol along with its associated data--modeling language, YANG, that significantly reduce network configuration complexity. However, NETCONF is limited to the interaction between managers and agents, and it has weak support for compliance to high-level management functionalities. We design and develop a network configuration management system called AutoConf that addresses the aforementioned problems. AutoConf is a distributed system that manages, validates, and automates the configuration of IP networks. We propose a new framework to augment NETCONF/YANG framework. This framework includes a Configuration Semantic Model (CSM), which provides a formal representation of domain knowledge needed to deploy a successful management system. Along with CSM, we develop a domain--specific language called Structured Configuration language to specify configuration tasks as well as high--level requirements. CSM/SCL together with NETCONF/YANG makes a powerful management system that supports network--wide configuration. AutoConf supports two levels of verifications: consistency verification and behavioral verification. We apply a set of logical formalizations to verifying the consistency and dependency of configuration parameters. In behavioral verification, we present a set of formal models and algorithms based on Binary Decision Diagram (BDD) to capture the behaviors of forwarding control lists that are deployed in firewalls, routers, and NAT devices. We also adopt an enhanced version of Dyna-Q algorithm to support dynamic adaptation of network configuration in response to changes occurred during network operation. This adaptation approach maintains a coherent relationship between high level requirements and low level device configuration. We evaluate AutoConf by running several configuration scenarios such as interface configuration, RIP configuration, OSPF configuration and MPLS configuration. We also evaluate AutoConf by running several simulation models to demonstrate the effectiveness and the scalability of handling large-scale networks