A comparison of overlay routing and multihoming route control Hayoung OH

2 Outline Routing Today BGP routing Overlay routing Overlay Routing vs. Multihoming Route Control Methodology of comparison Comparison results Discussion and summary

3 Routing Today ISPs perform wide-area routing through BGP Used to express local policy and traffic eng. Problem: users can’t express routing preferences Overlay routing Enables edge routing control Allows pooling of resources Problem: may interfere with ISP policy and traffic engineering

4 Routing Today :Example Default IP path determined by BGP Reroute traffic using red alternative overlay network path, avoid congestion point 1 Acts as overlay router

5 BGP Establish session on TCP port 179 Exchange all active routes Exchange incremental updates While connection is ALIVE exchange route UPDATE messages BGP = Border Gateway Protocol Is a Policy-Based routing protocol Is the de facto EGP of today’s global Internet Relatively simple protocol, but configuration is complex AS1 AS2 eBGP iBGP

6 BGP:Internet Routing BGP defines routes between stub networks UCLA Noho.net Berkeley.net UMass.net Internet 2 Mediaone C&W

7 BGP:Internet Routing BGP defines routes between stub networks UCLA Noho.net Berkeley.net UMass.net Internet 2 Mediaone C&W Noho-to-UMass

8 BGP:Internet Routing BGP defines routes between stub networks UCLA Noho.net Berkeley.net UMass.net Internet 2 Mediaone C&W Noho-to-Berkeley

9 BGP:Internet Routing UCLA Noho.net Berkeley.net UMass.net Internet 2 Mediaone C&W Noho-to-Berkeley Congestion or failure: Noho to Berkely BGP- determined route may not change (or will change slowly)

10 Overlay:Internet Routing UCLA Noho.net Berkeley.net UMass.net Internet 2 Mediaone C&W Noho-to-Berkeley Noho to UMass to Berkeley - route not visible or available via BGP! Congestion or failure: Noho to Berkely BGP- determined route may not change (or will change slowly) Overlay Routing Common characteristic:  Application-specific routing strategy Approach:  Ping and traceroute to learn underlying Internet topology

11 Overlay:RON (Resilient overlay networks) Discover good-quality paths through overlay nodes, and quickly turn to an alternate path when congestion or failure happens to the current path. Types of failures –Outages: Configuration/operational errors, backhoes, etc. –Performance failures: Severe congestion, denial-of-service attacks, etc. Scalable BGP-based IP routing substrate Reliability via path monitoring and re-routing Reliability via path monitoring and re-routing Reliability via path monitoring and re-routing Reliability via path monitoring and re-routing Measure all links between nodes Compute path properties Determine best route Forward traffic over that path

12 Overlay:RON design Prober Router Forwarder Conduit Performance Database Application-specific routing tables Policy routing module RON library Nodes in different routing domains (ASes) Layer 7 routing! application-layer router

13 BGP Inefficiencies Poor performance Policy routing, Coarse-grained route selection Poor recovery times BGP takes minutes to hours to recover

14 Example : Dual Rerouting Consider a native link failure in CE Only overlay link AE is affected. The native path AE is rerouted over F (ACE → ACFDE) Native Failure Overlay recovery: 8 Overlay rerouting: 4 Original: 2 Native Rerouting: 2 Time ∞ Native Recovery Native Repair Cost A B C D F E H I G A E I G OVERLAY NATIVE 4

15 Overlay Routing Bypass BGP routes end-to-end Flexible control on end-to-end path Improves performance Better recovery times

16 Number of Route Choices

17 Route selection Mechanism

18 Overlay Routing vs. Multihoming Route Control Is multihoming route control competitive with the flexibility of overlay routing systems? Yes  good performance and resilience achievable with BGP routing No  bypass mechanisms or changes to BGP may be necessary for improved performance and resilience

19 Talk outline Methodology of comparison Comparison results Discussion and summary

20 Comparison Methodology k-overlay performance depends on overlay size, node placement Results based on the testbed chosen

21 Measurement Testbed

22 Key Comparison Metrics Compare overlay and multihoming paths from nodes in a city to other nodes in the testbed.

23 Round-Trip Time Performance

24 Throughput Performance

25 Talk outline Methodology of comparison Comparison results Discussion and summary

26 RTT and Throughput Comparison

27 1-Overlays vs. k-Multihoming

28 1-Overlays vs. k-Multihoming

29 k-Overlays vs. k-Multihoming

30 k-Overlays vs. k-Multihoming

31 3-Overlays vs. 3-Multihoming

32 3-Overlays vs. 3-Multihoming

33 RTT and Throughput: Summary

34 Availability Comparison: Summary Use active ping measurements and RON failure data k-overlays offer almost perfect availability Multihoming may be necessary to avoid first-hop failures k-multihoming, k > 1, is not as perfect 3-multihoming: availability of 100% on 96% of city-dst pairs  1-multihoming: only 70% of pairs have 100% availability May be good enough for practical purposes

35 Talk outline Methodology of comparison Comparison results Discussion and summary

36 Overlay Routing vs. Multihoming Route Control

37 Conclusion Route control similar to overlay routing for most practical purposes Overlays may be overrated for end-to-end performance and resilience Don’t abandon BGP – there’s still hope

38 Reference A Comparison of Overlay Routing and Multihoming Route ControlA Comparison of Overlay Routing and Multihoming Route Control, Aditya Akella, SIGCOMM Route Flap Damping Exacerbates Internet Routing Convergence. Z.M.Mao, R.Govindan, G.Varghese,R.H.Kranz. SIGCOMM The Impact of Internet Policy and Topology on Delayed Routing Convergence. Craig Labovitz, Abha Ahuja, Roger Wattenhofer, Srinivasan Venkatachary. INFOCOM 2001