TIE Breaking: Tunable Interdomain Egress Selection Renata Teixeira Laboratoire d’Informatique de Paris 6 Université Pierre et Marie Curie with Tim Griffin.

TIE Breaking: Tunable Interdomain Egress Selection

Renata TeixeiraLaboratoire d’Informatique de Paris 6

Université Pierre et Marie Curie

with Tim Griffin (Cambridge), Mauricio Resende (AT&T), and Jennifer Rexford (Princeton)

2

Internet as a Communication Infrastructure

InternetHighly-sensitive to transient and persistent performance problems

3

Two-Tier Routing Architecture

InternetUCSD

Sprint

AT&T Verio

AOL

User

Web Server

UCSD

AT&T Verio

AOL

Interdomain routing (BGP)Selects AS-level path

based on policies

Intradomain routing (IGP)Most common: OSPF, IS-IS

Selects shortest path from ingress to egress based on link weights

4

UCSDSprint

AT&T Verio

AOL

Selecting Among Multiple Egresses Today

User

Web Server

UCSD

AT&T Verio

Hot-potato routingBGP selects closest egress by

comparing IGP distances

1 130

1 1525

NY

SF

LAB’s IGP distance

d(B,NY): 2d(B,SF): 31d(B,LA): 26

B

5

However, Hot-Potato Routing is…

Too disruptive Small changes inside can lead to big disruptions

A B

C

DG

EF4

5

11

39

34

108

68

A Bdst

Consequences-Transient forwarding instability-Traffic shift (largest traffic variations)-BGP updates to other domains

6

However, Hot-Potato Routing is…

Too disruptive Small changes inside can lead to big disruptions

Too restrictive Egress selection mechanism dictates a policy

Too convoluted IGP metrics determine BGP egress selection IGP paths and egress selection are coupled

7

Maybe a Fixed Ranking?

Goal: No disruptions because of internal changes Solution

Each router has a fixed ranking of egresses Select the highest-ranked egress for each destination Use tunnels from ingress to egress

Disadvantage Sometimes changing egresses would be useful

AB

C

DG

EF4

5

39

34

108

8

A Bdst

8

Egress Selection Mechanismsau

tom

atic

ada

ptat

ion

robustness to internal changes

hot-potato routing

fixed ranking

Explore trade-off

9

Metrics for Ranking Egresses

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

10

Goals of New Metric

Configurable Implement a wide-range of egress selection policies

Simple computation Compute on-line, in real-time Based on information already available in routers (distance)

Easy to optimize Expressive for a management system to optimize

Fine control Each ingress can implement its own ranking policy for each

destination

11

Decouples egress selection from internal paths Egress selection is done by tuning and

Allow a wide variety of egress selection policies Hot-potato: =1, = 0 Fixed ranking: =0, = constant rank

Requirements Small change in router decision logic Use of tunnels (as with fixed ranking)

m (e) = (e) . d(i,e) + (e)

TIE: Tunable Interdomain Egress Selection

mi(dst,e) = i(dst,e) . d(i,e) + i(dst,e)

weighted intradomain

distance

constant

12

Routers

ManagementSystem

Using TIE

Run optimization

, Configure routers

Path computationusing mi(dst,e)

Forwarding table

Administrator defines policy

Upon and changeor routing change

13

Configuring TIE to Minimize Sensitivity

Simulation Phase

Optimization Phase

Network topology

Set of egress routers per prefix

Set of failuressystem of

inequalities

Management System

configure routers withvalues i(dst,e) and i(dst,e)

that minimize sensitivity

14

At design time: mC(dst,A) < mC(dst,B)

9.C(dst,A) + C(dst,A) < 10.C(dst,B) + C(dst,B)

Simulation Phase

AB

C

dst

911

2010

20.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)

Output of simulation phase

C(dst,A)=1, C(dst,A)=1, C(dst,B) =2, C(dst,B) =0

15

Optimization Phase

One system of inequalities per (node, prefix) pair (num egresses – 1) x (num failures +1)

Practical requirements for setting parameters Finite-precision parameter values

Limiting the number of unique values

Robustness to unplanned events

Running time 37 seconds (Abilene network) and 12 minutes (ISP network)

• 196MHz MIPS R10000 processor on an SGI Challenge

Integer programming

Objective function: min (+)

1

16

Evaluation of TIE on Operational Networks

Topology and egress sets Abilene network (U.S. research network) Set link weight with geographic distance

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

Evaluation against hot-potato and fixed ranking Simulate single-node failures Measure routing sensitivity and delay

17

Sensitivity to Node Failures

fraction prefixes affected

CC

DF

of

(no

de,

fail

ure

) p

airs

15% of egress changes can be avoided without harming delay

18

Delay under Node FailuresC

CD

F o

f (

no

de,

des

tin

atio

n,f

ailu

re)

tup

les

ratio of delay after failure to design time delay

Under threshold, TIE has longer delay than hot-potato

It is better than fixed ranking for 60% of tuples

19

Conclusion

TIE mechanism for selecting egresses Decouples interdomain and intradomain routing Designed for being easy to optimize Small change to router implementation

Operators can optimize TIE for other policies Traffic engineering Robust traffic engineering Planning for maintenance

20

More details

http://rp.lip6.fr/~teixeira

21

UCSDSprint

AT&T Verio

AOL

Multiple Interdomain Egresses

User

Web Server

UCSD

AT&T Verio

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

NY

SF

LA

AOL

22

Why Hot-Potato Routing?

Independent and consistent egress decision Forward packet to neighbors that have selected

same (closest) egress

Minimize resource consumption Limits consumption of bandwidth by sending traffic

to next domain as early as possible

A B

C

DG

EF4

5

11

39

34

108

68

A Bdst

23

Summary ofBGP Decision Process

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

24

Other Policies

Traffic engineering Configure TIE parameters to select egresses to

obtain optimal link utilization Solution: Path-based multi-commodity flow

Robust traffic engineering Combine minimizing sensitivity with traffic

engineering problem

Preparing for maintenance

25

Traffic Engineering with TIE

Problem definition Balance utilization of internal links

Configure TIE parameters to select egresses to obtain optimal link utilization No need to set intradomain link weights

Solution Path-based multicommodity flow

No need to tweak routing protocols Avoid routing convergence

26

Example Policy: Minimizing Sensitivity

Problem definition Minimize sensitivity to equipment failures No delay more than twice design time delay

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

But cannot change routers for all possible goals

We can do this with TIE just by setting and