1 / 13

BGP ANYCAST Simulations Using GTNetS (work in progress)

BGP ANYCAST Simulations Using GTNetS (work in progress). Talal Jaafar Georgia Tech & CAIDA. Agenda. IP Anycasting Studies Our Approach GTNetS Simulations Experiments Remaining Work Q&A. IP Anycast Studies. Current measurements Planet Lab measurements (Sarat, Terzis et al)

ctorgerson
Download Presentation

BGP ANYCAST Simulations Using GTNetS (work in progress)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. BGP ANYCAST Simulations Using GTNetS (work in progress) Talal Jaafar Georgia Tech & CAIDA

  2. Agenda • IP Anycasting Studies • Our Approach • GTNetS Simulations • Experiments • Remaining Work • Q&A

  3. IP Anycast Studies • Current measurements • Planet Lab measurements (Sarat, Terzis et al) • Decrease in latency with anycast • Clients don’t always hit nearest server • Relatively small number of outages, but lasted for long time (>100s) • Hint more global nodes might cause instability • Ripe K-root measurements (Lorenzo) • Good latency • Local nodes take load off global nodes, but not by a great factor • Quite stable – few switches • Ballani and Francis with their Planet Lab measurements conclude that anycast nodes are quite stable – hardly any flips observed

  4. Why Simulations? • Study the impact of BGP on Anycast • BGP Convergence • Flap Dampening • Study the impact of Anycasting on BGP • BGP Churn • Large # of global nodes  Convergence impact? • Simulations might be useful in analyzing different options for good placements of future anycast servers

  5. GTNetS Simulation • Discrete-Event Packet Level simulations • BGP : BGP++ implementation ported from ns-2 (zebra based) • Simulation handles actual routing – longest prefix match based FIB which is populated by BGP • Anycast servers supported using /32 prefix address advertisement

  6. GTNetS Simulations (cont’d) • Failures • Silent link failure • Explicit withdraw of Anycast prefix • Modes of Deployment • Global nodes • Hierarchical Local-Global nodes • Hierarchical Local-Global nodes with Local nodesadvertising more specific prefix • Topology • Real AS Topology (RouteViews Project) • 2 Stages

  7. Tier-1 Topology Simulation • Representative f-root Internet backbone connections inferred from routeviews. • Simulate 1 BGP speaker per AS • Peer-Peer, Customer and Provider relations inferred and appropriate policies applied. • Total of 44 large ISPs with 467 interconnecting links simulated

  8. Tier-1 Topology Simulation (3) • Simulated Topology: • 10 Anycast servers (All Global) • 9 in North America • 1 in Europe • 34 clients • 22 in North America • 12 in Europe and Asia

  9. Tier-1 Topology Simulation Results Figure 1: Load distribution (44 Node Topology)

  10. Tier-1 Topology Simulation Results Figure 1: Geographic load distribution (44 Node Topology)

  11. Tier-1 and Tier-2 Topology Simulation • 5,476 Nodes and 14,468 Links • 12 Nodes provides service to F-Root • Distributed Simulation

  12. Research Goals • Use Simulations to: • Compare Unicast Vs. Anycast Routing • No failures • Link failures • Prefix withdrawals BGP Convergence BGP Churn Anycast Server Stability Anycast Load Balancing

  13. Questions ?

More Related