1. 1 An Experimental Analysis of BGP Convergence Time Timothy Griffin AT&T Research & Brian Premore Dartmouth College

2. 2 The problem There is a considerable delay in BGP convergence after a route change. The configurations on the autonomous system (AS) routers

3. 3 Solution By simulating a set of different router configurations, – find the optimal configuration in which the convergence time is minimal.

4. 4 Importance There is a considerable convergence delay after a route change Before the network converges – packet loss – packet delay – disruption of connectivity

5. 5 Roadmap BGP routing & reasons for long convergence time Some configurable options in BGP level routers Simulation network settings Analysis options vs. convergence time Summary

6. 6 BGP routing How BGP routing work  distance-vector algorithm Two types of update messages  Advertisements  withdrawals

7. 7 Long convergence time Nature of BGP path selection  Routers enumerate multiple paths  Eliminate and replacing other choices Minimum Route Advertisement Interval  MRAI, usually 30 seconds (hardcode)  Add delays Not guaranteed converge!!

8. 8 Roadmap BGP routing & reasons for long convergence time Some configurable options in BGP level routers Simulation network settings Analysis options vs. convergence time Summary

9. 9 Router configurations MRAI – Too high: longer delay between update message – Too low: oscillation of update message among routers  more update messages SSLD – Sender side loop detection WRATE – Withdrawal rate limiting

10. 10 Observations Depend on the network … – There is an optimal value of MRAI to minimize the number of update messages needed to be sent – There is another optimal value of MRAI to minimize the convergence time The affects of not using optimal value of MRAI – Very similar for all networks The affect of using SLLD and WRITE – Good and bad depends on the networks

11. 11 optimal values of MRAI number of update/withdrawal messages sent – MRAI small, oscillation – MRAI large … the convergence time – MRAI small … – MRAI large, longer to wait between messages

12. 12 using non-optimal MRAI Increase rapidly for MRAI < optimal Increase linearly for MRAI > optimal

13. 13 Roadmap BGP routing & reasons for long convergence time Some configurable options in BGP level routers Simulation network settings Analysis options vs. convergence time Summary

14. 14 Simulation setting SSFNet Java-based simulation package IP-level BGP-4 supported natively Networks Simple topology CLIQUE, CHAIN, RING, FOCUS, P-CLIQUE Limited size (5-20 nodes) Combinations of MRAI values, SSLD & WRATE Perl script, >200,000 runs, >1 week

15. 15 More on simulation setting One AS is one node in the network Only a single link update or withdrawal each time All routers have the same configurations No commercial extension Using network which guaranteed converge Random delays on processing route updates

16. 16 Roadmap BGP routing & reasons for long convergence time Some configurable options in BGP level routers Simulation network settings Analysis options vs. convergence time Summary

17. 17

18. 18 Roadmap BGP routing & reasons for long convergence time Some configurable options in BGP level routers Simulation network settings Analysis options vs. convergence time Summary

19. 19 Conclusion  We can configure the MRAI on BGP level routers to minimize convergence time.  The optimal value depends on the network  Having a higher MRAI value tends to be safer

20. 20 Future work  We know there is a optimal settings It depends on topology, link changes, … How can we configure the routers so the convergence time is close to the optimal value?