Improved BGP convergence via Ghost Flushing Yehuda Afek

Improved BGP convergence via Ghost Flushing Yehuda Afek Anat Bremler-Barr המרכז הבינתחומי הרצליה Shemer Schwarzd

Problem: BGP Convergence n n [Labovitz, Ahuja, Bose, Jahanian] BGP may take up to 15 minutes to converge. Here: Reduce the worst case from minutes to seconds, in a practical way

Problem: BGP Convergence n n n [Labovitz, Ahuja, Bose, Jahanian] Events Time (sec’s, min. Route. Adver=30) n 10, 000, up to 15 minutes n E-Down 30 • n n E-Up 30 • 30 d d 30, d=diameter n E-Longer 2 • 30 • l l == path length n E-Shorter 30 • 30 d Here: E-down = l time units (unit = link delay) E-Longer = 30 • d

Agenda n n n BGP overview The BGP convergence problem Ghost buster rule Ghost flushing rule Simulation results

BGP protocol n Distance (Path) vector protocol n Receive AS-path from the neighbors n Chooses the best one (shortest) n Eliminates Routing loops using the AS-path n Two kinds of messages: Announcements and Withdrawal

Problem: Ghost information One Ghost (old information) makes many, and in the network it continues recursively dst: 0 2 1 dst: 0 withdraw 0 dst 4 dst: 0 3 dst: 0 t=0

Problem: Ghost information One Ghost (old information) makes many, and in the network it continues recursively dst: 1 0 annc: 1 0 2 4 dst: 2 0 annc: 2 0 1 3 0 dst: 1 0 annc: 1 0 dst: {} t=1

Problem: Ghost information One Ghost (old information) makes many, and in the network it continues recursively dst: 3 1 0 2 4 dst: 1 2 0 dst: {} withdraw 1 3 dst: 1 2 0 0 dst: {} t=2

Problem: Ghost information One Ghost (old information) makes many, min. Route. Adver : Wait 30 sec’s before sending next announcement (BGP)recursively and in thethenetwork it continues dst: 3 1 0 annc: 3 1 0 dst: {} 2 4 dst: 2 1 0 annc: 2 1 0 1 3 dst: 2 1 0 annc: 2 1 0 0 dst: {} t=6 t=5 t=4 t=31 t=7 t=8 t=10 t=11 t=14 t=16 t=17 t=20 t=22 t=24 t=27 t=28

E_Down convergence BGP Time Message 30 n n. E In the clique (size 4) example the scenario ends after 62 sec (= 30(n-2) )

Without Min. Route. Adver n Avalanche of Messages O(n!) n Explore all possible paths of length 1, 2 … 1 : 1 0 dst: 0 2 3: 30 4: 40 1 2 : 2 0 dst: 0 3: 30 withdrawal 4: 40 0 dst 4 dst: 0 1: 10 2: 20 3: 30 3 dst: 0 1: 10 2: 20 4: 40 dst: {} t=0

Without Min. Route. Adver n Avalanche of Messages O(n!) n Explore all possible paths of length 1, 2 … 1 : 1 0 dst: 1 0 3 : 3 0 annc: 1 0 4: 40 dst: 2 0 2: 20 3 : 3 0 annc: 2 0 4: 40 2 4 1 3 0 dst: 1 0 1 : 1 0 annc: 1 0 3 : 3 0 4: 40 dst: {} t=0. 1

Without Min. Route. Adver n Avalanche of Messages O(n!) n Explore all possible paths of length 1, 2 … 1 : 1 2 0 dst: 3 0 3 : 3 0 annc: 3 0 4: 40 dst: 20 annc: 2 0 2: 20 3: 30 4: 40 2 4 1 3 0 dst: 2 0 1 : 1 2 0 annc: 2 0 2 : 2 0 3: 30 dst: 2 0 annc: 2 0 1 : 1 2 0 dst: {} 2: 20 4: 40 t=0. 2

Without Min. Route. Adver n Avalanche of Messages O(n!) n Explore all possible paths of length 2, 3 … 1 : 1 2 0 dst: 3 0 3: 30 annc: 3 0 4: 40 2 4 dst: 3 0 annc: 3 0 1 3 2: 230 3: 30 4: 40 0 dst: 3 0 1 : 1 2 0 annc: 3 0 2 : 2 1 0 3: 30 dst: 4 0 annc: 4 0 dst: {} t=0. 3 1: 120 2: 210 4: 40

Without Min. Route. Adver n Avalanche of Messages O(n!) n Explore all possible paths of length 2, 3 … 1 : 1 2 0 dst: 4 0 3: 310 annc: 4 0 4: 40 2 4 dst: 4 0 annc: 4 0 1 3 2: 230 3: 310 4: 40 0 dst: 1 2 0 1 : 1 2 0 annc: 1 2 0 2 : 2 1 0 3: 310 dst: 4 0 annc: 4 0 dst: {} t=0. 4 1: 120 2: 210 4: 40

E_Down convergence BGP with Min. Route. Adver BGP without Min. Route. Adver Time Message 30 n n. E hn n!E h=one link delay

Related Work n n n Introducing the problem [Labovitz, Ahuja, Bose, Jahanian], [Labovitz, Wattenhofer, Venkatachary, Ahuja] n real life evidence n theoretical analysis Experimental analysis [Griffin, Premore] Solution n Works in Counting to Infinity: n Adding states [Garcia-Luna-Aceves] – EIGRP like… n Route Poisoning with Hold-down [Cisco: Rutgers]– IGRP like. . . n Routes consistency [Pei, Zhao, Wang, Massey, Mankin, Wu, Zhang]

Ghost flushing rule n n If ASpath to dst is longer and cannot send annoucement (due to min. Route. Adver rule ) then send withdrawal Motivation: Flush the ghost information ASAP

Ghost Flushing example dst: 0 2 1 dst: 0 withdraw 0 dst 4 dst: 0 3 dst: 0 t=0

Ghost Flushing example dst: 1 0 annc: 1 0 2 4 dst: 2 0 annc: 2 0 1 3 0 dst: 1 0 annc: 1 0 dst: {} t=1

Ghost Flushing example Longer ASpath & min. Route. Adver timer Send “flushing” withdrawal dst: 3 1 0 2 4 dst: {} withdraw 1 3 withdraw 0 dst: 1 2 0 withdraw dst: {} t=2

Ghost Flushing example dst: {} withdraw 2 4 dst: {} 1 3 0 dst: {} withdraw dst: {} t=3

Analysis: Time convergence of ghost flushing rule, E_down n Ø Ø Ø In each time unit (=h, maximum link delay), ghost information is erased to a distance greater by one After k time units, ghost information ASpath with length < k has disappeared. Longest Ghost ASpath = n (in theory). Hence (worst case) time convergence: nh

E_Down convergence BGP with Min. Route. Adver BGP without Min. Route. Adver Ghost flushing Time Message 30 n n. E hn h=one link delay hn n!E 2 Ehn/30

Ghost Buster Rule n n The convergence time is better than expected !!!! Explanation: The min. Route. Adver blocks the propagation of ghost information, while the flushing withdrawal “eats” the ghost information. n Bad (wrong) news propagate slowly

Analysis: Ghost buster rule n Add to the ghost flushing rule: n n Router sends announcement, only after delta time Min. Route. Adver similar to n n delta: Common implementation: Min. Route. Adver peer And, timer almost always on (lots of BGP announcements !)

Analysis: Time convergence of ghost buster rule Ø The ghost information disappears at time t: d+t/(delta+h) = t/h Ø Ø Every delta+h time the length of the maximum ghost ASpath is increased by one. Every h time, the length of the minimum ghost ASpath is increased by one. After the failure the length of the maximum ghost ASpath is d (diameter). Hence: t = kdh/(k-1) d, where k = (delta+h)/h is the rate of the algorithm

E_Down convergence BGP with Min. Route. Adver BGP without Min. Route. Adver Ghost flushing With Ghost buster Time 30 n Message n. E hn n!E h=one link delay hn 2 Ehn/30 kdh/k-1 dh d=diameter k=(delta+h)/h 2 Ehkd/30(k-1)

The effect on E_longer n BGP: Convergence time dominated: 1. Time until ghost information vanishes 2. Time until backup path propagates in n Ghost flushing: helps the first factor 2 4 7 1 3 6 0 dst 5

The effect on E_longer n n Original BGP may err: n Min. Route. Adver peer stores wrong ASPath n BGP may err and send the packet in the wrong direction Ghost flushing: send withdrawal to a peer. Perhaps by a chance there may be an alternative path there.

Simulation: BGP code n Shortest path metric n Delay on link between 0. 2 to 2 sec n Min. Route. Adver randomly in 0 to 30 sec

Simulation: Clique E-down

Simulation: ISP topology 9 4 8 5 1 7 3 dst

Example: Core Internet (ASes) Out-degree In-degree BGP Ghost Flushing 1 45 10 963 22 2 52 17 898 51 3 3 4 1031 36 4 112 27 1017 50 5 61 11 1034 36 6 20 24 920 33 7 1 6 2 2. 5 8 18 2 1111 54 9 1 1111 981 62 10 1 98 4 5. 1

E_longer: Convergence Time 2 4 7 1 3 6 0 dst 5

Conclusion n Reduced convergence time from minutes to sec’s. n Does not hurt in other cases n Ghost flushing - no change at BGP messages n n Ghost buster solution – a new counting to infinity solution BGP very sensitive to minor modifications.