1. k-step Local Improvement Policy Motivation Inter-domain connectivity in the Internet is currently established on policy- based shortest-path routing. Business relationships of the Internet entities are translated into routing decisions through policies. Although these policies are built on simple mechanisms provided by BGP, they give rise to a very complex market structure. Extensive research efforts have been made to understand common practices of inter-domain policies. In early stages, hierarchical models were thought to be adequate to explain negotiation between the entities over improving routing and expressing routing preferences in general. Then, it has been realized that there are significant number of local policy exceptions and random decision making over the hierarchical structure. In this work, we examine how effective these local policy exceptions are in providing better quality paths. Our analysis on traces captured from the Internet quantitatively shows that currently adopted local policies could not be as effective as multi-hop negotiations for the purpose of attaining better paths in terms of multiple path quality metrics. Methodology Evaluation We define ``k-step Local Improvement Policy'‘ as an ISP attempts to: Establish paid peering with alternative providers Deal settlement free peering with other ISPs Broaden its geographic coverage via more points of presences to detour local performance bottlenecks and have wide-area route control. According to k-step local improvement policy, an ISP, is able to choose optimal paths within k-hop locality. otherwise, default best effort routing is applied beyond k-hop border. Our intention is to quantify overall benefits that can be ripened by these local policies against performance of multi-hop negotiation methods proposed by clean-slate architectures which offer end-to-end quality paths comparable to global optimum [4]. Relevant Publications [1] P. B. Godfrey, I. Ganichev, S. Shenker, and I. Stoica, “Pathlet routing,” in Proc. of SIGCOMM, 2009, pp. 111–122.. [2] H. V. Madhyastha, T. Anderson, A. Krishnamurthy, N. Spring, and A. Venkataramani, “A structural approach to latency prediction,” in Proc. of ACM SIGCOMM Internet Measurement Conference, 2006, pp. 99–104. [3] M. Yuksel, A. Gupta, and S. Kalyanaraman, “Contract-switching paradigm for internet value flows and risk management,” in Proc. of IEEE GI Symposium, 2008 [4] R. Mahajan, D. Wetherall, and T. Anderson, “Negotiation-based routing between neighboring isps,” in Proc. of USENIX NSDI, 2005 [5] H. T. Karaoglu and M. Yuksel, Inter-domain Multi-Hop Negotiation for the Internet in Proceedings of IEEE International Symposium on Policies for Distributed Systems and Networks (POLICY), Pisa, Italy, June 2011.. Current Improvements Methods An ISP can improve its connectivity via local methods: Direct peering with more ISPs (exchange points or private peering links) Broadening geographical coverage with more Points of Presence (PoP) Virtual Private Networks joined by a couple of ISPs Clean Slate Inter-domain Routing Proposals Multi-Hop Multi-Path Negotiation on Multiple Metrics Effectiveness of Current Routing Policies Bilateral Agreements – Due to bilateral nature of peering, an ISP has limited capability to influence route selection beyond its next hop. Single Metric and Single Path – Single metric based route calculation in its limitations may not always meet diverse performance metrics. A BE CF G 40ms, 1% 60ms, 0.1% 40ms, 1% 150ms, 0.5% 30ms, 1% 90ms, 2% 40ms, 0.5% ABEG: Least Packet Loss ACFG: Least Delay D 210ms, 3% ADG : Shortest Path In this work, we want to quantify increased effectiveness of these local policy improvement methods with broadening range of ISP control over selecting inter-domain routes via increased local collaboration with its neighbors in various levels. For best effort routing, we deploy a policy favoring Min AS Path, policy based shortest path routing and Closest Exit, hot potato routing predicts current Internet routes with 65% accuracy [2] We use iPlane PoP-level Internet topology [2] which provides latency and packet loss rate measurements for the links connecting ISPs' points of presence (PoP). In their work [2], authors have shown that these measurements have been proven stable through many repetitions. For our analysis, we followed these steps, randomly picked 101,000 PoP pairs. constructed end-to-end paths in according to (min AS Path, closest exit) policy k-hop Local Improvement policy with varying k values. selected the best possible path combining the segment within the k- hop locality with the path segment after k-hop border. compare performance improvement with best effort routing Latency Improvements Ex: k-step Local Improvement– Source ISP through peering agreements with its neighbors is able to choose optimal paths within k-PoP hops range. Beyond k-hop range, packets forwarded according to (min AS path, closest exit) method. Conclusions Maximum improvement, i.e., 70.16% for latency and 88.18% for packet loss, can be achieved via 13-step local improvement. Local improvement policies, i.e., direct peering and PoP establishment, usually extend to 2-hop locality can achieve significant performance gains up to 50%, especially higher for short paths become ineffective for path lengths beyond 6 PoPs. Number of ISPs need to be involved in end-to-end path negotiations which we call “diameter of negotiation” is approximately 5 on average. These results show us the necessity of automated SLA and signaling mechanisms in achieving desired multi-metric, multi-hop negotiation. Packet Loss Improvements Overall Improvements Diameter of Negotiation Destination Source This material is based upon work supported by the National Science Foundation under 0721600, 0721609, and 0831957. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation Inter-domain Multi-Hop Negotiation for the Internet Hasan T. Karaoglu (UNR) and Murat Yuksel (UNR) karaoglu@cse.unr.edukaraoglu@cse.unr.edu, yuksem@cse.unr.eduyuksem@cse.unr.edu Project Website: http://www.cse.unr.edu/~yuksem/contract-switching.htmhttp://www.cse.unr.edu/~yuksem/contract-switching.htm k-hop Local Improvement Policy Routing Multiple Performance Metric Cold Potato Routing Min Hop (Best Effort) Routing Min AS Path (only metric) Closest Exit