iLab Dynamic Routing Florian Wohlfart wohlfart@in.tum.de Chair of - - PowerPoint PPT Presentation

iLab

Dynamic Routing Florian Wohlfart wohlfart@in.tum.de

Chair of Network Architectures and Services Department of Informatics Technical University of Munich

Lab 3 – 18ss

1 / 42

Outline

Recap Background: Internet Architecture Internet Exchange Points (IXPs) Autonomous Systems Routing between ASes Theory: Routing Algorithms Problem Definition Link-State Routing Distance-Vector Routing Practice: Routing Prococols RIP OSPF BGP

2 / 42

Outline

Recap Background: Internet Architecture Internet Exchange Points (IXPs) Autonomous Systems Routing between ASes Theory: Routing Algorithms Problem Definition Link-State Routing Distance-Vector Routing Practice: Routing Prococols RIP OSPF BGP

3 / 42

Recap: Network Layer

wireless LAN browser TCP IP HTTP server TCP IP Ethernet driver WLAN driver IP Ethernet driver WLAN driver

HTTP protocol TCP protocol IP protocol IP protocol Ethernet protocol WLAN protocol

Ethernet router

4 / 42

Outline

Recap Background: Internet Architecture Internet Exchange Points (IXPs) Autonomous Systems Routing between ASes Theory: Routing Algorithms Problem Definition Link-State Routing Distance-Vector Routing Practice: Routing Prococols RIP OSPF BGP

5 / 42

Example: Traceroute from TUM I8 to google.de

LRZ DFN Google

◮ hops 1–4, Leibniz-Rechenzentrum (LRZ), AS12816 ◮ hops 5–6, Deutsches Forschungsnetz (DFN), AS680 ◮ hops 7–10, Google, AS15169 6 / 42

Example: DFN Network L2 Topology and PoPs

BAY BIR POT TUB GSI DUI BRE HAN BRA HEI REG DRE/DRN ZIB ILM JEN HUB ADH KEH BAS MUE KAI DES DKR KIE ROS BON KAS PAD GIE MAR GOE DOR WUP FHM WUE EWE AWI GRE FFO ZEU HAM

X-WiN-Topologie: Glasfasern

ERL CHE HWS FRE DUS KIT STU FRB FJL FZJ AAC AAH AUG Glasfaser Bestand Glasfaser geplant Kernnetzknoten Bestand Kernnetzknoten geplant

Stand: September 2015

SAA BIE MAG MDS LAP/LEI BOC FRA GAR

Source: https://www.dfn.de/fileadmin/1Dienstleistungen/XWIN/GF-Topo-2015-10-12.pdf

7 / 42

Internet Architecture

Internet Topology

◮ the Internet is a decentralized network of networks ◮ many organizations operate their own network, which all need

to be interconnected

◮ glue: IP protocol, exchange of routing information (BGP)

Typical Network Operators

◮ content providers

◮ content delivery networks (CDN) support content providers

◮ access providers (“eyeball networks”), content consumption ◮ transit providers 8 / 42

Internet Architecture cont’d

Source: C. Labovitz, S. Iekel-Johnson, D. McPherson, J. Oberheide, and F. Jahanian. Internet inter-domain traffic. In Proceedings of the ACM SIGCOMM 2010 conference (SIGCOMM ’10)

9 / 42

Example: Top Content Providers in Oregon Univ. Network

Source: B. Yeganeh, R. Rejaie, W. Willinger. A view from the edge: A stub-AS perspective of traffic localization and its implications. TMA 2017

10 / 42

Example Content Provider: The Google Edge Network

Source: Matt Calder et al. “Mapping the expansion of Google’s serving infrastructure” ACM IMC 2013.

◮ 15 data centers ◮ >100 Points of Presence (PoP)

◮ used for peering with other ASes at IXPs and private

interconnect facilities

◮ https://www.peeringdb.com/net/433

◮ >1000 Google Global Cache (GGC) inside ISP networks

◮ caching content close to end users Source: https://peering.google.com/#/infrastructure

11 / 42

Background: Internet Exchange Points (IXPs)

Definition

◮ an exchange point for traffic between network operators ◮ makes it easy to connect to many other networks (via peering) ◮ provides physical infrastructure

Largest IXPs

◮ DE-CIX (Frankfurt), see https://peeringdb.com/ix/31 ◮ AMS-IX (Amsterdam), see https://peeringdb.com/ix/26 ◮ LINX (London), see https://peeringdb.com/ix/18

Example fees (2016, DE-CIX Frankfurt)

◮ 1 Gbit/s costs 500 Euro/month ◮ 10 Gbit/s costs 1,550 Euro/month ◮ 100 Gbit/s costs 9,500 Euro/month 12 / 42

DE-CIX Frankfurt: Topology

Source: https://www.de-cix.net/about/topology/

13 / 42

DE-CIX Frankfurt: Core Switch

Source: https://press.de-cix.net/graphics/

14 / 42

DE-CIX Frankfurt: Throughput (2 days)

Source: https://www.de-cix.net/en/locations/germany/frankfurt/statistics

15 / 42

DE-CIX Frankfurt: Throughput (5 years)

Source: https://www.de-cix.net/en/locations/germany/frankfurt/statistics

16 / 42

Connecting Autonomous Systems

Definition: Autonomous System (AS)

◮ networks under a common administrative organization

e.g. ISP network, campus network

◮ an AS is identified by a 32bit AS-number

(was extended from 16bit in 2007)

◮ each AS advertises the address space it is willing to accept

(belonging to the AS itself or as transit traffic)

Implications

◮ from the outside, ASes are viewed as a single entity with border

routers and a routed address space

◮ changes within an AS are not relevant for other ASes

e.g. internal topology, intra-AS routing

◮ all ASes must agree on a common way of exchanging routing

information, inter-AS routing

17 / 42

Terminology: Autonomous Systems

18 / 42

Intra-AS ./ Inter-AS-Routing

Route selection

◮ intra-AS routing: focus on best paths ◮ inter-AS routing: also follow business policies

⇒ different link metrics and filtering policies

Scalability

border routers need to deal with a large number of routing table entries

◮ IPv4: >650k entries ◮ IPv6: 45k entries 19 / 42

Outline

Recap Background: Internet Architecture Internet Exchange Points (IXPs) Autonomous Systems Routing between ASes Theory: Routing Algorithms Problem Definition Link-State Routing Distance-Vector Routing Practice: Routing Prococols RIP OSPF BGP

20 / 42

Static /. Dynamic Routing

Static Routing

manual construction of routing tables

21 / 42

Static /. Dynamic Routing

Static Routing

manual construction of routing tables

◮ does not scale 21 / 42

Static /. Dynamic Routing

Static Routing

manual construction of routing tables

◮ does not scale ◮ does not react to changes (e.g. link failure) 21 / 42

Static /. Dynamic Routing

Static Routing

manual construction of routing tables

◮ does not scale ◮ does not react to changes (e.g. link failure)

Dynamic Routing

distributed algorithms automate the construction of routing tables

◮ scales depending on the algorithm ◮ automated routing table updates after topology changes (with

a certain delay)

21 / 42

Problem Definition

◮ find a path from the source to the destination host

D B C E F A

SRC DST

22 / 42

Problem Definition cont’d

◮ each hop (i.e. router or AS) forwards a packet closer to its

destination based on the information in its FIB SRC DST

23 / 42

Problem Definition cont’d

◮ link metrics attribute cost to links:

e.g. path length, reliability, delay, bandwidth, load, communication cost, or routing policies

D B C E F A

SRC DST

2 3 5 2 1 3 1 2 1

24 / 42

Problem Definition cont’d

routing can be reduced to a graph problem

◮ each node represents a router ◮ each edge represents a link ◮ each link comes with a certain cost, c(X,Y)

D B C E F A

2 3 5 2 1 3 1 2 1

25 / 42

Routing Protocols

Distributed routing algorithms perform the following steps

• 1. gather information about the network topology
• 2. create the local routing table based on the gathered information

Common Routing Protocols and Algorithms

◮ OSPF, IS-IS (Link-State Routing) ◮ RIP (Distance-Vector Routing) ◮ BGP (Path-Vector Routing) 26 / 42

Link-State Routing

Information propagation

◮ each router exchanges link-state updates with each other

router in the network (flooding)

◮ link-state updates contain list of adjacent routers and

networks including costs ⇒ each router knows the whole network topology (global view)

Information processing

◮ each router calculates the shortest paths to every other

destination, e.g. using Dijkstra’s algorithm

◮ derive routing table from shortest path tree

Implementations

◮ OSPF, IS-IS 27 / 42

Dijkstra’s Algorithm with Source Node u

I n i t i a l i z a t i o n : N’ = {u} # v i s i t e d nodes f o r a l l nodes v i f v i s neighbor

• f u then

D( v ) = c (u , v ) e l s e D( v ) = ∞ Loop f i n d w not in N’ such that D(w) i s minimum add w to N’ f o r each neighbor v

• f w

i f v not in N’ : D( v ) = min ( D( v ) , D(w) + c (w, v ) ) u n t i l N’ = N

28 / 42

Example: Dijkstra’s Algorithm for Router A

D B C E F A

2 3 5 2 1 3 1 2 1

N’ D(B), D(C), D(D), D(E), D(F), p(B) p(C) p(D) p(E) p(F)

29 / 42

Example: Dijkstra’s Algorithm for Router A

D B C E F A

2 3 5 2 1 3 1 2 1

N’ D(B), D(C), D(D), D(E), D(F), p(B) p(C) p(D) p(E) p(F) A 2,A ∞ 1,A ∞ ∞

29 / 42

Example: Dijkstra’s Algorithm for Router A

D B C E F A

2 3 5 2 1 3 1 2 1

N’ D(B), D(C), D(D), D(E), D(F), p(B) p(C) p(D) p(E) p(F) A 2,A ∞ 1,A ∞ ∞ A,D 2,A 4,D 2,D ∞

29 / 42

Example: Dijkstra’s Algorithm for Router A

D B C E F A

2 3 5 2 1 3 1 2 1

N’ D(B), D(C), D(D), D(E), D(F), p(B) p(C) p(D) p(E) p(F) A 2,A ∞ 1,A ∞ ∞ A,D 2,A 4,D 2,D ∞ A,D,E 2,A 3,E 4,E

29 / 42

Example: Dijkstra’s Algorithm for Router A

D B C E F A

2 3 5 2 1 3 1 2 1

N’ D(B), D(C), D(D), D(E), D(F), p(B) p(C) p(D) p(E) p(F) A 2,A ∞ 1,A ∞ ∞ A,D 2,A 4,D 2,D ∞ A,D,E 2,A 3,E 4,E A,D,E,B 3,E 4,E

29 / 42

Example: Dijkstra’s Algorithm for Router A

D B C E F A

2 3 5 2 1 3 1 2 1

N’ D(B), D(C), D(D), D(E), D(F), p(B) p(C) p(D) p(E) p(F) A 2,A ∞ 1,A ∞ ∞ A,D 2,A 4,D 2,D ∞ A,D,E 2,A 3,E 4,E A,D,E,B 3,E 4,E A,D,E,B,C 4,E

29 / 42

Example: Dijkstra’s Algorithm for Router A

D B C E F A

2 3 5 2 1 3 1 2 1

N’ D(B), D(C), D(D), D(E), D(F), p(B) p(C) p(D) p(E) p(F) A 2,A ∞ 1,A ∞ ∞ A,D 2,A 4,D 2,D ∞ A,D,E 2,A 3,E 4,E A,D,E,B 3,E 4,E A,D,E,B,C 4,E A,D,E,B,C,F

29 / 42

Distance-Vector Routing

Information propagation

◮ each router exchanges advertisements with its neighbors ◮ advertisements contain a snapshot of the current routing

table (as distance vectors) of the source router ⇒ the routers only have limited information about the network topology (local view)

Information processing

◮ each router keeps distance vectors to all other routers ◮ compute new distance vector on incoming advertisement ◮ derive routing table from distance vectors

Implementations

◮ RIP, BGP (path-vector) 30 / 42

Distance Vector Algorithm

Bellman-Ford equation:

Dx(y) = minv(c(x, v) + Dv(y)) minimum distance from x to y I n i t i a l i z a t i o n : f o r a l l nodes x , y in N Dx ( y ) = c ( x , y ) i f y i s neigbor

• f

x Dx ( y ) = ∞ e l s e Loop f o r e v e r : f o r a l l neighbors v

• f

x : Dx ( y ) = min v ( c ( x , v ) + Dv ( y )) i f Dx ( y ) changed : send Dx ( y ) to a l l neighbors

31 / 42

Example: Distance Vector Algorithm

X Y Z

7 2 1

node: X Y Z cost to: X Y Z X Y Z X Y Z from X: 2 7 ∞ ∞ ∞ ∞ ∞ ∞ from Y: ∞ ∞ ∞ 2 1 ∞ ∞ ∞ from Z: ∞ ∞ ∞ ∞ ∞ ∞ 7 1

32 / 42

Example: Distance Vector Algorithm

X Y Z

7 2 1

(0, 2, 7) (0, 2, 7)

node: X Y Z cost to: X Y Z X Y Z X Y Z from X: 2 7 2 7 2 7 from Y: ∞ ∞ ∞ 2 1 ∞ ∞ ∞ from Z: ∞ ∞ ∞ ∞ ∞ ∞ 7 1

32 / 42

Example: Distance Vector Algorithm

X Y Z

7 2 1

(2, 0, 1) (2, 0, 1)

node: X Y Z cost to: X Y Z X Y Z X Y Z from X: 2 3 2 7 2 7 from Y: 2 1 2 1 2 1 from Z: ∞ ∞ ∞ ∞ ∞ ∞ 3 1

32 / 42

Example: Distance Vector Algorithm

X Y Z

7 2 1

(3, 1, 0) (3, 1, 0)

node: X Y Z cost to: X Y Z X Y Z X Y Z from X: 2 3 2 7 2 7 from Y: 2 1 2 1 2 1 from Z: 3 1 3 1 3 1

32 / 42

Example: Distance Vector Algorithm

X Y Z

7 2 1

(0, 2, 3) (0, 2, 3)

node: X Y Z cost to: X Y Z X Y Z X Y Z from X: 2 3 2 3 2 3 from Y: 2 1 2 1 2 1 from Z: 3 1 3 1 3 1

32 / 42

Example: Distance Vector Algorithm

X Y Z

7 2 1

node: X Y Z cost to: X Y Z X Y Z X Y Z from X: 2 3 2 3 2 3 from Y: 2 1 2 1 2 1 from Z: 3 1 3 1 3 1

32 / 42

Example: Routing Table Computation

node: X Y Z cost to: X Y Z X Y Z X Y Z from X: 2 3 2 3 2 3 from Y: 2 1 2 1 2 1 from Z: 3 1 3 1 3 1 From which router was the best route learned? node: X Y Z cost to: X Y Z X Y Z X Y Z via X: 2 4 5 7 9 10 via Y: 4 2 3 3 1 2 via Z: 10 8 7 4 2 1 e.g. routing table of router X: networks announced by router Y: next hop Y (cost 2), networks announced by router Z: next hop Y (cost 3)

33 / 42

Example: Count-to-Infinity Problem

X Y Z

7 50 1

node: Y Z cost to: X Y Z X Y Z from X: 2 3 2 3 from Y: 4 1 2 1 from Z: 3 1 3 1

34 / 42

Example: Count-to-Infinity Problem

X Y Z

7 50 1

(4, 0, 1) (4, 0, 1)

node: Y Z cost to: X Y Z X Y Z from X: 2 3 2 3 from Y: 4 1 4 1 from Z: 3 1 5 1

34 / 42

Example: Count-to-Infinity Problem

X Y Z

7 50 1

(5, 1, 0) (5, 1, 0)

node: Y Z cost to: X Y Z X Y Z from X: 2 3 2 3 from Y: 6 1 4 1 from Z: 5 1 5 1

34 / 42

Example: Count-to-Infinity Problem

X Y Z

7 50 1

(6, 0, 1) (6, 0, 1)

node: Y Z cost to: X Y Z X Y Z from X: 2 3 2 3 from Y: 6 1 6 1 from Z: 5 1 7 1

34 / 42

Count-to-Infinity Countermeasures

Distance-Vector with Poisoned Reverse

◮ if the shortest path to a destination Y was learned from

neighbor Z, then set DX(Y ) = ∞ in updates to neighbor Z

Path-Vector Routing

◮ Update messages contain full path, this allows loop-detection ◮ e.g. ([0,B], [4,"D,E,A"], [1,"E,F"]) 35 / 42

Example: Poisoned Reverse

X Y Z

7 50 1

node: Y Z cost to: X Y Z X Y Z from X: 2 3 2 3 from Y: 4 1 2 1 from Z: 3 1 3 1

36 / 42

Example: Poisoned Reverse

X Y Z

7 50 1

(∞, 0, 1) (∞, 0, 1)

node: Y Z cost to: X Y Z X Y Z from X: 2 3 2 3 from Y: 4 1 7 1 from Z: 3 1 ∞ 1

36 / 42

Example: Poisoned Reverse

X Y Z

7 50 1

(7, 1, 0) (7, 1, 0)

node: Y Z cost to: X Y Z X Y Z from X: 2 3 2 3 from Y: 8 1 7 1 from Z: 7 1 ∞ 1

36 / 42

Distance-Vector (DV) ./ Link-State Routing (LS)

Scalibility

◮ LS floods the network ◮ DV nodes only talk to their neighbors

Speed of convergence

◮ LS converges fast ◮ DV can converge slowly, e.g. count-to-infinity problem

routing loops may occur temporarily

37 / 42

Outline

Recap Background: Internet Architecture Internet Exchange Points (IXPs) Autonomous Systems Routing between ASes Theory: Routing Algorithms Problem Definition Link-State Routing Distance-Vector Routing Practice: Routing Prococols RIP OSPF BGP

38 / 42

Routing Information Protocol (RIP)

◮ RIPv2, RIPng (supports IPv6)

Basics

◮ interior gateway protocol (IGP) ◮ distance-vector algorithm ◮ link metric: number of hops (link cost= 1, ∞ = 16) ◮ implements poisoned reverse

Routing updates

◮ UDP port 520 ◮ RIP response messages are sent periodically (every 30s) or on

changes

◮ link is declared unreachable after 180s without an update ◮ each advertisement contains routes to max. 25 destinations 39 / 42

Open Shortest Path First (OSPF)

◮ conceived as the successor to RIP, some advanced features ◮ OSPFv3 adds support for IPv6 (2008)

Basics

◮ interior gateway protocol (IGP) ◮ link-state algorithm ◮ multiple same-cost paths ◮ supports hierarchy through definition of areas ◮ messages can be authenticated

Routing updates

◮ IP datagrams with protocol number 89, multicast addressing ◮ can be authenticated 40 / 42

Border Gateway Protocol (BGP)

Background

◮ BGP is the de facto inter-AS routing protocol in the Internet ◮ BGPv4 was introduced in 1994 ◮ extensible to implement new functions ◮ built to scale ◮ implements policy-based routing

Basics

◮ path-vector algorithm (path is a sequece of AS numbers) ◮ uses periodic keepalives and incremental updates 41 / 42

Border Gateway Protocol (BGP) cont’d.

Routing updates

◮ TCP port 179 ◮ eBGP spreads inter-AS routing information between ASes ◮ iBGP spreads inter-AS routing information within an AS ◮ path is a sequence of AS numbers

Problems

◮ routing table growth (>650.000 IPv4 entries) ◮ load balancing in multihomed ASes ◮ security: BGP prefix hijacking

e.g. https://www.ripe.net/publications/news/industry-developments/youtube-hijacking-a-ripe-ncc-ris-case- study

42 / 42