TIE Breaking Tunable Interdomain Egress Selection Renata Teixeira

TIE Breaking: Tunable Interdomain Egress Selection Renata Teixeira Laboratoire d’Informatique de Paris 6 Université Pierre et Marie Curie with Tim Griffin (Cambridge), Mauricio Resende (AT&T), and Jennifer Rexford (Princeton)

Internet as a Communication Infrastructure Internet Highly-sensitive to transient and persistent performance problems 2

Two-Tier Routing Architecture Web Server AT&T Verio Internet UCSD User Interdomain routing (BGP) Selects AS-level path based on policies AOL Sprint Intradomain routing (IGP) Most common: OSPF, IS-IS Selects shortest path from ingress to egress based on link weights 3

Selecting Among Multiple Egresses Today Web Server AT&T Verio UCSD User SF Sprint NY 1 1 B 30 1 1 5 AOL 25 LA B’s IGP distance d(B, NY): 2 d(B, SF): 31 d(B, LA): 26 Hot-potato routing BGP selects closest egress by comparing IGP distances 4

However, Hot-Potato Routing is… s Too disruptive § Small changes inside can lead to big disruptions dst A 4 D 3 11 F 5 3 8 C B 9 E 8 10 4 G 6 Consequences -Transient forwarding instability -Traffic shift (largest traffic variations) -BGP updates to other domains 5

However, Hot-Potato Routing is… s Too disruptive § Small changes inside can lead to big disruptions s Too restrictive § Egress selection mechanism dictates a policy s Too convoluted § IGP metrics determine BGP egress selection § IGP paths and egress selection are coupled 6

Maybe a Fixed Ranking? s Goal: No disruptions because of internal changes s Solution § Each router has a fixed ranking of egresses § Select the highest-ranked egress for each destination § Use tunnels from ingress to egress dst A 4 s Disadvantage 3 F 5 B 9 D 3 8 8 10 E 4 G C § Sometimes changing egresses would be useful 7

automatic adaptation Egress Selection Mechanisms hot-potato routing Explore trade-off fixed ranking robustness to internal changes 8

Metrics for Ranking Egresses s Egress selection mechanisms are based on a metric (m) that each ingress router (i) uses to rank each egress router (e) for a destination § Hot-potato routing • m is the intradomain distance (d(i, e)) § Fixed ranking • m is a constant 9

Goals of New Metric s Configurable § Implement a wide-range of egress selection policies s Simple computation § Compute on-line, in real-time § Based on information already available in routers (distance) s Easy to optimize § Expressive for a management system to optimize s Fine control § Each ingress can implement its own ranking policy for each destination 10

TIE: Tunable Interdomain Egress Selection mi(dst, e) (e) == (e) i(dst, e). d(i, e) + i(dst, e) weightedfrom internal constant s Decouples egress selection paths § Egress selection isintradomain done by tuning and distance s Allow a wide variety of egress selection policies § Hot-potato: =1, = 0 § Fixed ranking: =0, = constant rank s Requirements § Small change in router decision logic § Use of tunnels (as with fixed ranking) 11

Using TIE Administrator defines policy Management System Routers Run optimization , Configure routers Upon and change or routing change Path computation using mi(dst, e) Forwarding table 12

Configuring TIE to Minimize Sensitivity Network topology Set of egress routers per prefix Set of failures Management System Simulation Phase system of inequalities Optimization Phase configure routers with values i(dst, e) and i(dst, e) that minimize sensitivity 13

Simulation Phase dst B A 20 9 11 10 At design time: C m. C(dst, A) < m. C(dst, B) Output of simulation phase 9. C(dst, A) + C(dst, A) < 10. C(dst, B) + C(dst, B) 11. C(dst, A) + C(dst, A) < 10. C(dst, B) + C(dst, B) 20. C(dst, A) + C(dst, A) > 10. C(dst, B) + C(dst, B) C(dst, A)=1, C(dst, B) =2, C(dst, B) =0 14

Optimization Phase s One system of inequalities per (node, prefix) pair § (num egresses – 1) x (num failures +1) s Practical requirements for setting parameters § Finite-precision parameter values Integer programming § Limiting the number of unique values Objective function: min ( + ) § Robustness to unplanned events 1 s Running time § 37 seconds (Abilene network) and 12 minutes (ISP network) • 196 MHz MIPS R 10000 processor on an SGI Challenge 15

Evaluation of TIE on Operational Networks s Topology and egress sets § Abilene network (U. S. research network) § Set link weight with geographic distance s Configuration of TIE § § Considering single-link failures Threshold of delay ratio: 2 [1, 4] and 93% of i(dst, e)=1 {0, 1, 3251} and 90% of i(dst, e)=0 s Evaluation against hot-potato and fixed ranking § Simulate single-node failures § Measure routing sensitivity and delay 16

CCDF of (node, failure) pairs Sensitivity to Node Failures 15% of egress changes can be avoided without harming delay fraction prefixes affected 17

CCDF of (node, destination, failure) tuples Delay under Node Failures It is better than fixed ranking for 60% of tuples Under threshold, TIE has longer delay than hot-potato ratio of delay after failure to design time delay 18

Conclusion s TIE mechanism for selecting egresses § Decouples interdomain and intradomain routing § Designed for being easy to optimize § Small change to router implementation s Operators can optimize TIE for other policies § Traffic engineering § Robust traffic engineering § Planning for maintenance 19

More details http: //rp. lip 6. fr/~teixeira 20

Multiple Interdomain Egresses Web Server AT&T Verio UCSD User NY SF Sprint AOL LA Multiple egresses for a destination are common! ISPs usually peer in multiple locations and customers buy multiple connections to one or more ISPs for reliability and performance 21

Why Hot-Potato Routing? s Independent and consistent egress decision § Forward packet to neighbors that have selected same (closest) egress s Minimize resource consumption § Limits consumption of bandwidth by sending traffic to next domain as early as possible dst A 4 D 3 11 F 5 3 8 C B 9 E 8 10 4 G 6 22

Summary of BGP Decision Process s BGP decision process § § § § Ignore if exit point unreachable Highest local preference Lowest AS path length Lowest origin type Lowest MED (with same next hop AS) Lowest IGP cost to next hop Lowest router ID of BGP speaker 23

Other Policies s Traffic engineering § Configure TIE parameters to select egresses to obtain optimal link utilization § Solution: Path-based multi-commodity flow s Robust traffic engineering § Combine minimizing sensitivity with traffic engineering problem s Preparing for maintenance 24

Traffic Engineering with TIE s Problem definition § Balance utilization of internal links s Configure TIE parameters to select egresses to obtain optimal link utilization § No need to set intradomain link weights s Solution § Path-based multicommodity flow s No need to tweak routing protocols § Avoid routing convergence 25

Example Policy: Minimizing Sensitivity s Problem definition § Minimize sensitivity to equipment failures § No delay more than twice design time delay s Would be a simple change to routers § If distance is more than twice original distance • Change to closest egress § Else • Keep using old egress point s But cannot change routers for all possible goals We can do this with TIE just by setting and 26