MIRO: M Multi ulti- -path path MIRO: Interdomain nterdomain RO ROuting uting I Wen Xu and Jennifer Rexford Princeton University 1 Overlay Entertainment 1 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 1 Ming
Outline Outline � Motivation � Routing Architectures � MIRO Protocol Design � Performance Evaluation � Conclusion 2 Overlay Entertainment 2 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 2 Ming
If Some is Good, More is Better If Some is Good, More is Better � More flexible path selection � Avoid a path with an undesirable hop � Better load balancing � Split traffic over multiple paths � Faster recovery from failure � Switch from one path another � Secure routing � Avoid paths controlled by an adversary 3 Overlay Entertainment 3 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 3 Ming
Outline Outline � Motivation � Routing Architectures � MIRO Protocol Design � Performance Evaluation � Conclusion 4 Overlay Entertainment 4 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 4 Ming
Overlay Network Overlay Network C B A F D E � Relay traffic through an intermidate node � Not good for reasons � Increase latency , bandwidth consumption of relays � Probing overhead 5 Overlay Entertainment 5 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 5 Ming
Source Routing Source Routing C B ABCF ABEF A F ADECF D E � Sender determines paths for packets � Difficult challenges � Scalability � Limited control for intermediate ASes 6 Overlay Entertainment 6 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 6 Ming
Border Gateway Protocol Border Gateway Protocol You could reach 1.2.3.0/ 20 You could reach through me Prefix AS path 1.2.3.0/ 20 I have C through me 1.2.3.0/ 20 1.2.3.0/ 20 A B C F B A D E F 1.2.3.0/20 A doesn’t pay for A F traffic transit D E � Policy-based routing � AS Relationships BGP session Route propagation � Provider-customer Outbound filter � Peer-peer 7 Overlay Entertainment 7 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 7 Ming
BGP Routing Process BGP Routing Process Routing I nformation Base Best Route Selection 1. Highest local-pref 2. Shortest AS path SI NGLE path routing protocol I nbound filters Outbound filters 3. Lowest origin type 4. Lowest MED 5. eBGP over iBGP 6. Lowest I GP cost 7. Tie-break Best Route BGP routing table 8 Overlay Entertainment 8 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 8 Ming
Outline Outline � Motivation � Routing Architectures � MIRO Protocol Design � Performance Evaluation � Conclusion 9 Overlay Entertainment 9 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 9 Ming
Maybe BGP isn’ ’t So Bad t So Bad… … Maybe BGP isn � Internet is a big, federated network � Local policies for path selection & export routes � Efficient propagation of path information � Add multi-path to BGP � AS-level, path-vector routing (more scalable) � Extended for multi-path 10 Overlay Entertainment 10 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 10 Ming
Pull- -based Negotiation based Negotiation Pull CF* CEF Any route to F BEF* CBEF How about avoiding E? BCF BCF is OK BCF? C B A F ABEF* ADEF D E EF* DEF* ECF DABEF � Pull-based route retrieval � Solicit routes only when necessary � Bilateral negotiation � AS relationships usually bilateral anyway 11 Overlay Entertainment 11 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 11 Ming
Not Just Your Neighbors Not Just Your Neighbors CF* CEF * BEF CBEF * BCF YES BCF C Any route to F B avoiding EF? BCF is OK A F ABCF* ABEF* ADEF D E EF* DEF* ECF DABEF DABCF � Multi-party, multi-hop � Either AS can initiate the negotiation 12 Overlay Entertainment 12 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 12 Ming
Tunnels for Forwarding Data Packets Tunnels for Forwarding Data Packets B’s incoming CF* CEF tunnel table BEF* CBEF BCF BCF is OK 7 A-> F BCF F C B F F F Tunnel id = 7 A F ABEF* ADEF D E A’s outgoing tunnel table EF* DEF* ECF F B, 7 BCF DABEF � Direct packet along alternate route � Destination-based forwarding not enough � Encapsulate the packet to egress point 13 Overlay Entertainment 13 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 13 Ming
Advantages Advantages � Flexibility for edge AS � AS can learn and use multiple paths � Control for intermediate ASes � Selective export of extra routes � Scalability of the global system � AS-level routing protocol (extension to BGP) � Pull-based route retrieval � Selective export of routes 14 Overlay Entertainment 14 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 14 Ming
Outline Outline � Motivation � Routing Architectures � MIRO Protocol Design � Performance Evaluation � Conclusion 15 Overlay Entertainment 15 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 15 Ming
Infer AS Graph & Relationships Infer AS Graph & Relationships Observed paths provider-customer peer-peer A B E F A D E F C B B C F C E F A F D E � RouteViews BGP data in 2000, 2003, and 2005 � Gao inference algorithm 16 Overlay Entertainment 16 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 16 Ming
Three Variations on Export Rules Three Variations on Export Rules C B A F D E provider-customer peer-peer provider route � Strict(/s): all paths in the same class peer route customer route � Export(/e): all policy-compliant paths � Flexible(/a): all of the paths, relationship independent 17 Overlay Entertainment 17 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 17 Ming
Evaluation Methodology Evaluation Methodology � Who to negotiate with? � Neighboring ASes � ASes along the default path � How many ASes initiate negotiation? � Only sending AS � What path-selection policies? � Avoid a particular AS in default path 18 Overlay Entertainment 18 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 18 Ming
Exposing the Underlying Path Diversity Exposing the Underlying Path Diversity > 100 alternates > 10 alternates 5% have no alternate paths 19 Overlay Entertainment 19 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 19 Ming
Avoiding an AS: Success Ratio Success Ratio Avoiding an AS: Date BGP MIRO/s MIRO/e MIRO/a Source Routing 2000 27.8% 65.4% 72.9% 75.3% 89.5% 2003 31.2% 67.0% 74.6% 76.6% 90.4% 2005 29.5% 67.8% 73.7% 76.0% 91.1% MI RO is much better Source routing only than BGP 10-15% better 20 Overlay Entertainment 20 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 20 Ming
Avoiding an AS: Overhead (for 2005) Avoiding an AS: Overhead (for 2005) Policy Success Rate AS# /tuple Path# /tuple Strict 67.8% 2.80 36.6 Export 73.7% 2.53 58.9 Flexible 76.0% 2.38 139.0 � With more flexible policies � Negotiate with fewer ASes � But learn more paths from each AS 21 Overlay Entertainment 21 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 21 Ming
Incremental Deployment Incremental Deployment 100 Percentage of total gain nodes with more than 200 neighbors 80 60 Even if only a few ASes adopt MI RO, nodes with degree 40 it’s still effective! greater than 40 20 0 0.01 0.1 1 10 100 Percentage of ASes Adopted MI RO (log scale) 2005/ s 2005/ e 2005/ a 22 Overlay Entertainment 22 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 22 Ming
Outline Outline � Motivation � Routing Architectures � MIRO Protocol Design � Performance Evaluation � Conclusion 23 Overlay Entertainment 23 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 23 Ming
Conclusion Conclusion � Multi-path extension to BGP � Retain AS-level, path-vector routing � Add pull-based route retrieval � Selective export of extra routes � Tunnels for forwarding data packets � Evaluation results � Satisfies the AS’s policy goals � Avoids state explosion � Offers benefits in small deployments 24 Overlay Entertainment 24 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 24 Ming
Reference Reference � Slides from Wen Xu (many thanks!) 25 Overlay Entertainment 25 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 25 Ming
Happy Entertainment Happy Entertainment Thank you~ kero kero kero 26 Overlay Entertainment 26 Ming- -Tsang Tsai Tsang Tsai Overlay Entertainment 26 Ming
Recommend
More recommend
