Improving internet resilience through lightweight preventive detection (lpd) and persistent detection and recovery (pdr)
Lan Wang, Lixia Zhang · 2004
It is important for the Internet to function well under adverse conditions caused by operator errors, hardware failures, software flaws, and malicious attacks. As the Internet continues to grow, however, this goal becomes more difficult to achieve since network protocols often need to maintain more state and there are potentially more sources of faults and attacks that can introduce false state information and/or state inconsistencies. In this dissertation, we propose two efficient approaches to improving network protocols' resilience against network faults and attacks. The first approach, Lightweight Preventive Detection (LPD), exploits network and protocol characteristics to identify protocol messages containing false state information. Because it does not rely on complex cryptographic mechanisms, this approach can produce fast detection mechanisms that are readily deployable in the current Internet. The second approach, Persistent Detection & Recovery (PDR), allows network nodes to periodically correct state inconsistencies without incurring per-state refresh overhead. The key idea behind this approach is state compression—every node compresses its entire state space into a digest and the digest is used to detect and recover inconsistent state. We apply the two proposed approaches to protecting BGP (Border Gateway Protocol) against falsely injected routes and undetected routing inconsistencies. Our Adaptive Path-Filtering mechanism exploits routing stability and server redundancy to protect important Internet server sites, such as the top-level DNS (Domain Name System) servers, against route-hyjacking. Our Fast Routing Table Recovery (FRTR) mechanism allows BGP to recover efficiently from routing table corruptions as well as transient session failures. Our trace-driven simulation shows that (1) adaptive path-filtering effectively filter out false routes to top-level DNS servers, and it has little negative impact on DNS service accessibility; and (2) FRTR achieves error-free routing tables with a probability close to 100% after only a few rounds of message exchanges, and it reduces the bandwidth overhead of session recovery by one to two orders of magnitude. Furthermore, we apply the PDR approach to RSVP (ReSerVation Protocol). Our Scalable Refreshes mechanism can effectively maintain the state consistency among RSVP nodes while incurring only a constant refresh overhead. Finally, we propose and evaluate two generic state compression techniques, Digest Tree and Bloom Filter-based Digest, which can be incorporated into any Persistent Detection & Recovery mechanism. The effectiveness of these protocol mechanisms and techniques demonstrates that our approaches can lead to a resilient protocol design.