Since its early days, the Internet has witnessed not only a phenomenal growth, but also a large number of security attacks, and in recent years, denial-of-service (DoS) attacks have emerged as one of the top threats. The stateless and destination-oriented Internet routing combined with the ability to harness a large number of compromised machines and the relative ease and low costs of launching such attacks has made this a hard problem to address. Additionally, the myriad requirements of scalability, incremental deployment, adequate user privacy protections, and appropriate economic incentives has further complicated the design of DDoS defense mechanisms. While the many research proposals to date have focussed differently on prevention, mitigation, or traceback of DDoS attacks, the lack of a comprehensive approach satisfying the different design criteria for successful attack attribution is indeed disturbing.

Our first contribution here has been the design of a composable data model that has helped us represent the various dimensions of the attack attribution problem, particularly the performance attributes of accuracy, effectiveness, speed and overhead, as orthogonal and mutually independent design considerations. We have then designed custom optimizations along each of these dimensions, and have further integrated them into a single composite model, to provide strong performance guarantees. Thus, the proposed model has given us a single framework that can not only address the individual shortcomings of the various known attack attribution techniques, but also provide a more wholesome counter-measure against DDoS attacks.

Our second contribution here has been a concrete implementation based on the proposed composable data model, having adopted a graph-theoretic approach to identify and subsequently stitch together individual edge fragments in the Internet graph to reveal the true routing path of any network data packet. The proposed approach has been analyzed through theoretical and experimental evaluation across multiple metrics, including scalability, incremental deployment, speed and efficiency of the distributed algorithm, and finally the total overhead associated with its deployment. We have thereby shown that it is realistically feasible to provide strong performance and scalability guarantees for Internet-wide attack attribution.

Our third contribution here has further advanced the state of the art by directly identifying individual path fragments in the Internet graph, having adopted a distributed divide-and-conquer approach employing simple recurrence relations as individual building blocks. A detailed analysis of the proposed approach on real-life Internet topologies with respect to network storage and traffic overhead, has provided a more realistic characterization. Thus, not only does the proposed approach lend well for simplified operations at scale but can also provide robust network-wide performance and security guarantees for Internet-wide attack attribution.

Our final contribution here has introduced the notion of anonymity in the overall attack attribution process to significantly broaden its scope. The highly invasive nature of wide-spread data gathering for network traceback continues to violate one of the key principles of Internet use today - the ability to stay anonymous and operate freely without retribution. In this regard, we have successfully reconciled these mutually divergent requirements to make it not only economically feasible and politically viable but also socially acceptable.

This work opens up several directions for future research - analysis of existing attack attribution techniques to identify further scope for improvements, incorporation of newer attributes into the design framework of the composable data model abstraction, and finally design of newer attack attribution techniques that comprehensively integrate the various attack prevention, mitigation and traceback techniques in an efficient manner.