R outing - PowerPoint Presentation

Slide1

R

outing

State Distance:A Path-based Metric forNetwork Analysis

Natali RuchanskyGonca Gürsun, Evimaria Terzi, and Mark CrovellaSlide2

Shortest Path Distance

Distance Metrics for Analyzing Routing

2

Similarly RoutedSlide3

Based on this distance intuition we develop a new metric based on paths and show it is good for:Visualization

of networks and routesCharacterizing routesDetecting significant patternsGaining insight about routing

A New Metric3Slide4

We call this path-based distance metric:Routing State

Distance4Slide5

Conceptually…

Imagine capturing the entire interdomain routing state of the internet in a matrix

the next hop on path from

to Each row is the routing table of a single ASNow consider the columns…

 5

Sources

DestinationsSlide6

We define between two prefixes

and as the number of entries that differ in

their columns of  

Routing State Distance6i.e. the number of ASes

that disagree about the next-hop to and

.

 Slide7

More FormallyGiven a universe

of prefixes define:A next-hop matrix :

the next-hop on the path to

As well as :

7Slide8

RSD to BGPIn order to apply to measured BGP paths

we define to have ASes on rows and prefixes on columns.

the next-hop from AS to prefix

 8

Solution Key: is defined on a set

of

paths

NOT a graph

A few issues arise…

Missing Values

Multiple next hopsSlide9

Our DataFrom 48 million AS paths consisting of:

359 unique monitors450K destination prefixes We end up with:

243 sources ASes 130K prefixes

Thus our is

 9Slide10

Why is appealing? 

Let’s take a look at its properties…10Slide11

RSD versus Hop Distance11No relation between RSD and hop distanceSlide12

Finer Grained MeasureVaries smoothly and has a

gradual slope. Allows fine granularity

12

Increase of 1 encompasses many prefixesSlide13

Highly structured Allows 2D visualization

13

From compute

, our distance matrix where: 

 Slide14

Wow

! Highly structured14

This happens with

any

random sample  Internet-wideSlide15

Yeah, but a cluster of what!?!Now in routing

terms:Any row in

must have the same next hop in nearly each cellThe set of ASes

make similar routing decisions w.r.t destinations  

First think matrix-wise ():A cluster

corresponds to a set of columns Columns being close in means they are

similar

in some positions

is highly

coherent

15

We call such a pair

a

local atom

 Slide16

16

Small cluster “C”

Large Cluster

Small cluster “C”

Large clusterSlide17

A local atom is a set of prefixes that are routed similarly in some region of the internet.

So the smaller cluster is a local atom of certain prefixes

that are routed similarly by a large set of ASes

17Slide18

For this investigate … Prefer a specific AS for transit to these prefixes. Hurricane Electric (HE)

If any path passes through HE :Source

ASes prefer that pathPrefix appears in the smaller cluster

. Why these specific prefixes?

Level3

Hurricane Electric

Sprint

18Slide19

But why do sources always route through HE if the option exists?

….HE has a relatively unique peering policy.Offer peering to

ANY AS with presence in the same exchange point.HE’s peers prefer using

HE for ANY customer of HEAnd hence consists of networks that peer with HE,

and consists of HE’s customers

 19Slide20

Analysis with uncovered a macroscopic atom

Can we formulate a systematic study to uncover other smaller atoms?Intuitively

we would like a partitioning of the prefixes such that :In the same

group is minimizedBetween different groups is maximized  Can We Find More Clusters?

20Slide21

RS-Clustering ProblemIntuition: A

partitioning of the prefixes such that :In the same group is minimizedBetween

different groups is maximized For a partition

:

Key Advantage

:

Parameter Free!

21Slide22

Optimal is HardFinding the optimal solution to the

Problem is NP-hardWe propose two approaches

:Pivot ClusteringOverlap Clustering

 22Slide23

Given a set of prefixes , their

values, and a threshold parameter :Start from a random prefix

(the pivot)Find all

that fall within distance to and form a clusterRemove cluster from

and repeatAdvantages:The algorithm is fast

: O(|E|)Provable approximation guarantee 

Pivot Clustering Algorithm

23Slide24

5 largest clusters

Clusters show a clear separationEach cluster corresponds to a local atom24Slide25

25

Size of C

Size of S

Destinations

C1

150

16

Ukraine 83%

Czech. Rep 10%

C2

170

9

Romania

33%

Poland 33%

C3

126

7

India 93%

US 2%

C4

484

8

Russia 73%

Czech

rep. 10%

C5

375

15

US 74%

Australia 16%

Interpreting ClustersSlide26

To address this we propose a formalism called Overlap Clustering and show that it is capable of extracting such clusters.

We ask ourselves if a

partition is really best?

26

Seek a clustering that captures

overlapSlide27

Related WorkReported that BGP tables provide an incomplete view of the AS graph. [

Roughan et. al. ‘11]Visualization based on AS degree and geo-location. [Huffaker

and k. claffy ‘10]Small scale visualization

through BGPlay and bgpvizClustering on the inferred AS graph. [Gkantsidis et. al. ‘03]Grouping prefixes that share the same BGP paths into policy atoms

. [Broido and k. claffy

‘01]Methods for calculating policy atoms and characteristics. [Afek et. al. ‘02]

27Slide28

Take-AwayAnalysis with typical distance metrics is hardWe introduce a new one -- Routing

State Distance – that is simple and based only on pathsOvercome BGP hurdles and

show it can be used for:In-depth analysis of BGPCapturing closeness useful for visualizationUncovering surprising

patternsGeneral settingDeveloped a new set of tools forextracting insight from BGP measurements28Slide29

Code, data, and more information is available on our website at:

csr.bu.edu/rsd29

Code

Pivot Clustering

Overlap Clustering

RSD Computation

Data

Prefix List

Pairwise RSDSlide30

Natali

RuchanskyGonca Gürsun, Evimaria

Terzi, and Mark Crovella

Thank you!!