This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Internet Routing (COS Internet Routing (COS 598A)598A)
• Disadvantage: changes the answers– Clients only learn a subset of the BGP routes– Does not result is same choices as a full
mesh– ... especially if RR sees different IGP
distances
Routing Anomaly: Forwarding Loop
r1 r2
1
1
1r1r2
Picks r2 Picks r1
Packet deflected toward other egress point, causing a loop
Routing Anomaly: Protocol Oscillation
5 5 5
r1 r2 r3
1 1 11 2 3
RR1 prefers r2 over r1RR2 prefers r3 over r2RR3 prefers r1 over r3
Avoiding Routing Anomalies
• Reduce impact of route reflectors– Ensure route reflector is close to its clients– … so the RR makes consistent decisions
• Sufficient conditions for ensuring consistency– RR preferring routes through clients over “peers”– BGP messages should traverse same path as data
• Forces a high degree of replication– Many route reflectors in the network– E.g., a route reflector per PoP for correctness– E.g. have a second RR per PoP for reliability
• Make route reflectors more verbose– Send all BGP routes to clients, not just best route– Send all equally-good BGP routes (up to IGP cost)
• Advantages– Client routers have improved visibility– Make the same decisions as in a full mesh
• Disadvantages– Higher overhead for sending and storing routes– Requires protocol changes to send multiple routes– Not backwards compatible with legacy routers
• Replace point-to-point distribution– Apply a multicast protocol to distribute
messages– Or, flood the BGP messages to all routers
• Advantages– Complete distribution without route reflectors– Avoids configuration overhead of a full mesh
• Disadvantages– Requires an additional, new protocol– Not backwards-compatible with legacy routers– Large BGP routing tables, like in a full mesh
http://www.nanog.org/mtg-0302/ppt/van.pdf
Possible Solution: Tunnel Between Edge Routers
r1 r2
1
1
1r1r2tunneltunnel
• Tunneling through the core– Ingress router selects ingress point– Other routers blindly forward to the egress
• Advantages– No risk of forwarding loops– No BGP running on interior routers
• Disadvantages– Overhead of tunneling protocol/technology– Still has a risk of protocol oscillations
State-of-the-Art of BGP Distribution in an AS
• When full-mesh doesn’t scale– Hierarchical route-reflector configuration
• One or two route reflectors per PoP
– Some networks use “confederations” (mini ASes)• Recent ideas
– Sufficient conditions to avoid anomalies– Enhanced RRs sending multiple or custom
routes– Flooding/multicast of BGP updates– Tunneling to avoid packet deflections
• Open questions– Are the sufficient conditions too restrictive?– Good comparison of the various approaches
IGP Topology
Interior Gateway Protocols (IGPs)
• Protocol overhead depends on the topology– Bandwidth: flooding of link state advertisements – Memory: storing the link-state database– Processing: computing the shortest paths
32
2
1
13
1
4
5
3
Improving the Scaling
• Dijkstra’s shortest-path algorithm– Simplest version: O(N2), where N is # of nodes– Better algorithms: O(L*log(N)), where L is #
links– Incremental algorithms: great for small
changes
• Timers to pace operations– Minimum time between LSAs for the same link– Minimum time between path computations
• More resources on the routers– Routers with more CPU and memory
Introducing Hierarchy: OSPF Areas
• Divide network into regions– Backbone (area 0) and non-backbone areas– Each area has its own link-state database– Advertise only path distances at area boundaries
Area 0
Area 1 Area 2
Area 3 Area 4
areaborderrouter
32
2
1
13
1
4
5
3
To area 0
cost 3
cost 8
Summarization at Area Boundaries
• Areas only help so much– Advertising path costs to reach each component– Single link failure may change multiple path costs
• Summarization: LSA for multiple components– LSA for an IP prefix containing the addresses– LSA carries cost for the maximum path cost
32
2
1
13
1
4
5
3
To area 0
cost 8
Assigning OSPF Areas
• Group related routers– E.g., in a Point-of-Presence– Assign to single OSPF area– Put inter-PoP links in area 0
• Enable summarization– Select an address block for
the equipment in the area– Assign IP addresses in the
block to router CPUs and interfaces
Intra-PoP
Other networks
Inter-PoP
Pros and Cons of Summarization
• Advantages: scalability– Reduce the size of the link-state database
• One entry per summary prefix
– Isolate the rest of the network from changes• Only advertise when max path cost changes
• Disadvantages– Complexity
• Extra configuration details for areas & summarization
• Requires tight coupling with IP address assignment
– Inefficiency• Summarization hides details that affect path
selection• Data packets may traverse a less-attractive path
Dividing into Multiple ASes
• Divide the network into regions– Separate instance of IGP per region– Interdomain routing between regions– Loss of visibility into differences within region
100
100
100
100
100
10020 20 20 20 20 20
50
50 50
50
50
50
North America Europe Asia
Multi-AS Networks, Not Just for Scalability
• Administrative reasons– Separate networks per geographic region– Mergers/acquisitions that combine networks
• Why not merge to single AS?– Using different intradomain protocols– Managed by different people– Fear of encountering scalability problems– Fear of losing the benefits of isolation
• Why merge to a single AS?– Simpler configuration– More efficient routing– Avoid having separate AS hop in BGP AS paths
Which Approach is Better?
• Ideal: flat IGP network– Single AS– Single IGP instance, no areas
• Hierarchical IGP– Single AS– Single IGP instance, using areas &
summarization
• Multiple ASes– Multiple ASes– Separate IGP instances
• Some other approach???
Comparison Metrics
• Scalability– Protocol overhead
• Storing and flooding link-state advertisements• Overhead of Dijkstra shortest-path computation
– Effects of topology changes• Number of advertisements after a change• Likelihood a change must be propagated
• Efficiency– Stretch: comparing path lengths
• In ideal flat intradomain routing• In alternative scheme
– How much longer do the paths get?
Interesting Research Questions
• Routing protocols that achieve small stretch– Theory work on algorithms to minimize stretch– Protocol work on hierarchy and aggregation– Any new distributed protocols with low stretch?– Avoid sharp boundaries between areas/ASes?
• Identifying good places to hide information– Given a network graph with link weights– Decide where to put area and AS boundaries– … with the goal of minimizing stretch– … within some max size of each area or AS
Conclusion
• Networks are getting bigger– Growth of a network topology– Merger/acquisition of other networks
• Techniques for scaling the routing design– BGP route reflection– OSPF areas– Multiple BGP ASes
• Relatively open research area– Rich theoretical tradition on compact routing– Common operational practices for protocol
scaling– Not much work has been done in between
Next Time: Router Configuration
• Two papers– “Automated provisioning of BGP customers”
(just sections 1-3)– “Detecting BGP faults with static analysis”
• Review only of second paper– Summary– Why accept– Why reject– Future work
• Optional– Short survey on BGP routing policies for ISPs– NANOG video covering material in second