INL: IP Networking Lab Université Catholique de Louvain The Benefits of Locator/ID Separation AGAVE: A lightweight Approach for Viable End-to-end IP-based QoS Services Joint Work: L. Iannone, O.Bonaventure, B. Quoitin, D. Saucez, C. De Launois 1
Roadmap Motivation Loc/ID Separation How does it work Benefits Improved QoS FIB Reduction Costs 2
Roadmap Motivation Loc/ID Separation How does it work Benefits Improved QoS FIB Reduction Costs 3
Multihoming => Path diversity AS j Internet Default Free Zone (DFZ) AS w ASx AS z AS k ASy 4
Multihoming => Path diversity AS j Internet AS w ASx AS z AS k ASy 4
Multihoming => Path diversity Path-Vector nature of BGP does not allow full usage of Path Diversity AS j Internet AS w ASx AS z AS k ASy 4
BGP table growth Main causes: Active BGP entries (FIB) PI (Provider Independent) prefix assignement Multi-homing Traffic-Engineering... Date 01-Jul-89 to 21-Sep-07 Source: http://www.cidr-report.org 5
Roadmap Motivation Loc/ID Separation How does it work Benefits Improved QoS FIB Reduction Costs 6
Separating Locators and IDs AS j Internet AS w ASx AS z AS k ASy 7
Separating Locators and IDs AS j Internet AS w ASx AS z AS k ASy 7
Separating Locators and IDs AS j Internet AS w RLOC 2 RLOC 1 ASx AS z AS k ASy RLOC 1 RLOC 2 7
Separating Locators and IDs AS j Internet AS w RLOC 2 RLOC 1 ASx AS z AS k ASy RLOC 1 RLOC 2 7
Separating Locators and IDs AS j Internet AS w RLOC 2 RLOC 1 ASx AS z AS k ASy RLOC 1 RLOC 2 7
Separating Locators and IDs Mapping Distribution System? AS j Internet AS w RLOC 2 RLOC 1 AS z AS k ASx ASy RLOC 1 RLOC 2 7
Separating Locators and IDs Mapping Distribution System RLOC 2 RLOC 1 AS j Internet AS w RLOC 2 RLOC 1 AS z AS k ASx ASy RLOC 1 RLOC 2 7
Separating Locators and IDs AS j Internet AS w RLOC 2 RLOC 1 AS z AS k ASx ASy RLOC 1 RLOC 2 RLOC 2 RLOC 2 7
Separating Locators and IDs AS j Internet AS w RLOC 2 RLOC 1 AS z AS k ASx ASy RLOC 1 RLOC 2 RLOC 2 RLOC 2 7
Separating Locators and IDs AS j Internet AS w RLOC 2 RLOC 1 AS z AS k ASx ASy RLOC 1 RLOC 2 RLOC 2 RLOC 2 7
Separating Locators and IDs AS j Internet AS w RLOC 2 RLOC 1 AS z AS k ASx ASy RLOC 1 RLOC 2 RLOC 2 RLOC 2 7
Roadmap Motivation Loc/ID Separation How does it work Benefits Improved QoS FIB Reduction Costs 8
Advantage 2: Path diversity RLOC 2 ASx AS j Playing with RLOCs enables using path diversity for QoS AS z AS k AS w RLOC 1 ASy RLOC 1 RLOC 2 9
Internet Path-diversity Number of of destination prefixes 30 k 25 k 20 k 15 k 10 k 5 k Using BGP routes Using LISP Routeviews Dataset Simulations of Multihomed ASes 0 k 0 2 4 6 8 10 12 14 16 18 20 22 24 Number of available paths Dataset Routes Peers Pairs M-h Stubs M-h Prefixes Routeviews 5,750,380 32 496 6,402 29,575 RIPE NCC00 1,641,618 11 55 6,247 29,934 10
Exploiting Path-diversity 450 400 Delay : lowest, median, worst 350 Delay (ms) [ms] 300 250 200 150 100 50 0 0 20 40 60 80 100 120 140 160 Pair Pair of of multihomed sites sites Simulations of 13 multi-homed ASes (156 pairs) Based on RIPE NCC Traffic Measurements Service Delays averaged using the Vivaldi algorithm How to take advantage of path diversity: TSC (Tunnel Service Controller) to select end-points 11
Exploiting Path-diversity 450 400 350 Delay : lowest, median, worst Pair 45: Delay (ms) [ms] 300 250 200 150 100 50 0 0 20 40 60 80 100 120 140 160 Pair Pair of of multihomed sites sites Simulations of 13 multi-homed ASes (156 pairs) Based on RIPE NCC Traffic Measurements Service Delays averaged using the Vivaldi algorithm How to take advantage of path diversity: TSC (Tunnel Service Controller) to select end-points 11
Exploiting Path-diversity 450 400 350 Delay : lowest, median, worst Pair 45: Best Path: ~100 ms Delay (ms) [ms] 300 250 200 150 100 50 0 0 20 40 60 80 100 120 140 160 Pair Pair of of multihomed sites sites Simulations of 13 multi-homed ASes (156 pairs) Based on RIPE NCC Traffic Measurements Service Delays averaged using the Vivaldi algorithm How to take advantage of path diversity: TSC (Tunnel Service Controller) to select end-points 11
Exploiting Path-diversity Delay (ms) [ms] 450 400 350 300 250 200 150 100 50 Delay : lowest, median, worst Pair 45: Best Path: ~100 ms Worst Path: ~300 ms 0 0 20 40 60 80 100 120 140 160 Pair Pair of of multihomed sites sites Simulations of 13 multi-homed ASes (156 pairs) Based on RIPE NCC Traffic Measurements Service Delays averaged using the Vivaldi algorithm How to take advantage of path diversity: TSC (Tunnel Service Controller) to select end-points 11
Exploiting Path-diversity Delay (ms) [ms] 450 400 350 300 250 200 150 100 50 Delay : lowest, median, worst Pair 45: Best Path: ~100 ms Worst Path: ~300 ms Median: ~ 150 ms 0 0 20 40 60 80 100 120 140 160 Pair Pair of of multihomed sites sites Simulations of 13 multi-homed ASes (156 pairs) Based on RIPE NCC Traffic Measurements Service Delays averaged using the Vivaldi algorithm How to take advantage of path diversity: TSC (Tunnel Service Controller) to select end-points 11
Roadmap Motivation Loc/ID Separation How does it work Benefits Improved QoS FIB Reduction Costs 12
Advantage 1: shrinking the FIBs AS j Less Prefixes to be Globally announced AS w ASx AS z AS k ASy 13
Shrinking the FIB (I) Default Free Zone (DFZ) Tier-1 (PI Prefixes) AS A j AS B w Transit AS A.1 k A.2 AS B.1 k AS B.2 k Stub A.1.1 A.2.1 B.1.1 Separating address space allows new strategies for assigning routing locators More topologically driven Less Provider Independent (PI) prefixes 14
Shrinking the FIB (II) P(X)>x Simulations Synthetic topologies generated with GHITTLE Hierarchical with business relationship 14965 ASes Number of FIB entries (logscale) 15
Shrinking the FIB (II) P(X)>x Simulations Synthetic topologies generated with GHITTLE Hierarchical with business relationship 14965 ASes Number of FIB entries (logscale) 15
Cool, but... There Ain't No Such Thing As A Free Lunch! 16
Roadmap Motivation Loc/ID Separation How does it work Benefits Improved QoS FIB Reduction Costs 17
Netflow data collection Full Netflow (v7) on border router 1 Gigabit link to Belnet ~10000 users (/16 prefix block) Analysis: flow-tools + custom software /BGP Granularity of mappings iplane data set 18
Mappings Cache Size 110000 99000 88000 Number of Entries 77000 66000 55000 44000 33000 22000 11000 0 00h 12h 24h 3 Min Timeout 30 Min Timeout 300 Min Timeout 19
Lookups 3 min timeout Timeout Period 1 RLOC 2 RLOCs 3 RLOCs 3 min. Night Day 10.17 kbps 34.97 kpbs 12.17 kbps 41.85 kbps 14.17 kbps 48.74 kbps 30 min. Night Day 2.04 kbps 8.95 kbps 2.44 kbps 10.71 kbps 2.84 kbps 12.47 kbps 300 min. Night Day 0.163 kbps 2.68 kbps 0.195 kbps 3.21 kbps 0.227 kbps 3.74 kbps 20
Conclusions Loc/ID Separation approach allows: Exploiting Path Diversity Reduce FIB size Loc/ID Separation costs: Cache Lookups & Tunneling overhead 21
Thanks?? /* */ http://inl.info.ucl.ac.be http://www.ist-agave.org 22