AS Assignment for Routers Bradley Huffaker bradley@caida.org Amogh - PowerPoint PPT Presentation

AS Assignment for Routers Bradley Huffaker bradley@caida.org Amogh Dhamdhere, Marina Fomenkov, kc claffy CAIDA University of California at San Diego, La Jolla, CA AIMS Workshop -- February 2010

Overview • motivation • methodology • analysis • conclusions 2

AS Assignment Problem motivation Which AS, 32 or 12, owns/controls the router a ? ? AS 32 AS 12 120.8.10.23 23.13.32.2 a 120.8.10.23 23.13.32.2 IP address prefix 120.8.10.0/24 23.13.0.0/16 AS 32 12 router ? BGP longest matching prefix BGP origin AS As assignment 23.13.32.2 23.13.0.0/16 12 ? 120.8.10.23 120.8.10.0/24 32 3

Motivation motivation • Dual graph - a combined router and AS graph • Dual graph analysis - Relationship between AS degree and the AS’s number of routers. - how does heuristic assignment affect the inferred number of routers in an AS • More accurate AS traceroute - resolving AS loops 4

Here is What We Want motivation Dual Router and AS graph 1 2 3 4 5

This is What We Have motivation Router graph with interfaces. 9.0.1.1 10.0.2.3 10.0.1.5 5.5.1.28 10.0.1.1 13.5.1.8 6

Mapping to Prefix motivation Router graph with prefixes assigned to links. 9.0.1.1 10.0.2.3 9.0.1.0/24 10.0.0.0/16 10.0.1.5 5.5.1.28 10.0.1.1 10.0.1.0/25 5.5.1.0/24 10.0.1.0/24 13.5.1.8 13.5.1.0/24 7

Mapping to ASes motivation Router graph with AS assigned to links. 9.0.1.0/24 1 10.0.0.0/16 2 2 10.0.1.0/25 5.5.1.0/24 10.0.1.0/24 2 3 13.5.1.0/24 4 8

Assigning AS to Routers motivation Router graph with AS assigned to routers. 1 2 1 3 2 2 2 2 2 2 3 4 4 9

Dual Graph motivation 1 1 2 2 3 3 2 2 2 4 4 10

Methodology methodology We compared the success rates of four different AS assignment heuristics against our ground truth data sets. 11

Ground Truth methodology • ISPs (i) - Tier 1, Tier 2, and five research networks • interface sets - I i interfaces in the address space of ISP i , on routers that do belong to ISP i - I i interfaces in the address space of ISP i on routers that do not belong to ISP i • router sets - R i is the set of routers with interfaces in the address space of ISP i that do belong to ISP i - R i is the set of routers with interfaces in the address space of ISP i that do not belong to ISP i • AS sets - A i is the set of ASes that do belong to ISP i - A i is the set of ASes that do not belong to ISP i 12

Ground Truth methodology R R routers owned routers not owned Tier I f,h 3,405 2,254 Tier 2 h 241 86 GEANT f 37 0 I-Light f 32 0 Internet 2 f 17 0 National LambdaRail f 16 0 CANET f 8 0 f Organization provided f ull interface list h Organization provided naming h euristic 13 that allowed for inference of R

Data sources methodology • Router Graph (MAARS 1 ) - Sept. - Oct. 2009 - 268 million traceroute paths - 22 million nodes 2 / 22 million links 3 • BGP Data - Oct. 2009 - 311,230 prefixes • AS relationships - Oct. 2009 - BGP data - 148,565 AS relationship pairs 1 router alias resolver 2 node = set of IPs on same router 14 3 link can connect > 2 nodes

Data Topology methodology 4 address space Interface sets as 2 color 9.0.2.10 I 12 10.0.1.1, 10.0.2.3, 10.0.1.6 c c 2 I 12 10.0.1.28 d 4 d 9.0.1.1 12 10.0.2.3 router sets 7 R 12 b, d, f 12 R 12 a a b b e 10.0.1.5 10.0.1.16 e AS sets 10.0.1.1 10.0.1.28 A 12 12 A 12 4, 2, 7 we assume a has a f uninferred interface f 13.5.1.8 7 route AS type which does not belong a 12 single-AS to 12 b 4, 12 multi-AS c 4 single-AS b gets candidate AS from d 2, 12 multi-AS its interface 10.0.1.1 and e 12 single-AS the link it shares with c . f 12, 7 multi-AS f has no interface in I 12 and I 12 , so has no known 15 ownership

AS assignment methods methodology Single Single : only one choice A provider Customer Election : most interfaces B A A A - more links into router’s ISP’s address B Election space A A A A A A customer Neighbor : most single AS neighbors B A A - connected to more routers owned by the router’s ISP C Degree Customer : customer AS AS DEGREE Neighbor - customer’s router uses provider’s A 1 A A address space for the interconnect B 2 A C 3 A C A A Degree : smallest degree AS B C D C C A A A - proxy for Customer, large degree AS D B B typically is provider of small degree A A B B AS B 16

Methodology methodology • primary method - assignment is used if it is not ambiguous • tie-breaker method - method with highest success rate on routers for which primary method yields ambiguous results ambiguous election no majority AS among links neighbor no majority AS among neighbors customer no unambiguous customer relationship among ASes degree tie between smallest degree ASes 17

counting success? methodology successful assignment : If router r is known to be owned by ISP i and method H(r) selects an AS owned by ISP i , or if r is known to not be owned by an ISP i and method H(r) selects an AS not owned by ISP i . 18

Method Success Rates analysis 100 S - success rate R tie-breaker Election + Degree performs R tie-breaker F - failure rate R primary best with 80% success rate. 80 R primary 60 Tier 1 bias in ground truth reduces precentage accuracy of customer and degree heuristics 40 Tie-breaker ambiguous assignments not counted 20 0 S F S F S F S F S F Customer Degree All Election Neighbor + + + + Degree Degree Neighbor Neighbor 19 single AS ------------------multiple AS--------------- 72% 28%

Success Rates analysis • single AS routers - all methods successful for R (67% of single AS routers) - all methods fail for R ( 33% of single AS routers) routers in R i must have an interface in A i , therefore single AS routers only have an AS in A i , making it impossible for any method to select an AS in A i . I i is ISP i ’s address space so it maps to A i . failed to find or resolve real router X ownership is not known, so is discarded interface alias X I i I i I i A i A i R i R i ? ? X A i X 20

Success Rates analysis • multiple AS routers (28%) - Election + Degree best with 80% success rate. • single AS routers (72%) - all methods successful for R (67% of single AS routers) - all methods fail for R ( 33% of single AS routers) • overall - Election + Degree best with 70% success rate. 21

Analysis of Dual Topology analysis number of single AS 1e+07 1e+06 100000 routers 10000 1000 100 10 1 1 10 100 1000 10000 AS degree 100000 of single AS routers Median number 10000 1000 100 10 1 1 10 100 1000 AS degree statistical correlation that we can use for topology scaling and generation 22

Heuristic Effect on AS Router Count analysis how do different heuristics affect number of inferred routers per AS 100 Election Customer Median number of multi-AS routers Neighbor Degree Neighbor assigns more nodes to large degree ASes 10 Customer assigns more nodes to small degree ASes 1 1 10 100 1000 AS degree 23

Resolving AS Loops analysis interface/link path C A B D D A A C A B packet received on D, but response sent from A 24

Resolving AS Loops analysis interface/link path C A B D D A A C A B packet received on D, but response sent from A router path Using inferred AS assignments resolves apparent AS loop. D C A B D C A B D A D C A B 24

Resolved AS Loops analysis 0.65 Neighbor resolved the fraction of traceroute AS path loops resolved 0.6 most loops with 63%. 0.55 Election+Degree (the 0.5 combination with the 0.45 greatest success rate) resolves 62% of AS loops 0.4 0.35 1~5% of paths contain AS 0.3 loops, depending on the monitor. 0.25 0.2 Election Customer Neighbor Degree Election+ Degree 25

Conclusion conclusion • multiple AS routers - Election + Degree best with 80% success rate. • all routers - Election + Degree best with 70% success rate. • AS loop resolution - Election+Degree resolves 62% or AS loops 26

Future Work/What we need future work • More ground truth • alternative AS assignment heuristics Bradley Huffaker bradley@caida.org http://www.caida.org/publications/papers/2010/as_assignment/ 27