![Page 1: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/1.jpg)
High PerformanceSwitching and RoutingTelecom Center Workshop: Sept 4, 1997.
Eliminating Packet Loss Caused by BGP Convergence
Nate Kushman
Srikanth Kandula, Dina Katabi, and Bruce Maggs
![Page 2: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/2.jpg)
BGP Convergence Causes You Packet Loss
• Route changes cause up to 30% packet loss for more than 2 minutes [Labovitz00]
• Even for domains dual homed to tier 1 providers, a failover event can cause multiple loss bursts, and one loss burst can last for up to 20s [Wang06]
• Popular and unpopular prefixes experience losses due to BGP convergence [Wang05]
• 50% of VoIP disruptions are highly correlated with BGP updates [Kushman06]
The Problem:
![Page 3: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/3.jpg)
What Kind of Solution Do We Want?
• Eliminate packet loss during BGP convergence
• An adopting ISP protects itself and its customers from loss even if no other ISP cooperates
![Page 4: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/4.jpg)
Transient Path Loss Problem
Nobody offers Patrick an alternate pathNobody offers Patrick an alternate path MIT
AT&T
Avi
Patrick
Sprint
All of Patrick’s providers are using him to get to MIT
![Page 5: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/5.jpg)
Transient Path Loss Problem
MIT
AT&T
Avi
Patrick
Sprint
Link DownLOSS!
Patrick knows no alternate path to MIT
Patrick drops AT&T’s and Avi’s packets to MIT, and his own
![Page 6: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/6.jpg)
Transient Path Loss Problem
MIT
AT&T
Avi
Patrick
Sprint
Link Down
Eventually, Patrick withdraws path from AT&T and Avi
AT&T and Avi stop sending packets to Patrick
![Page 7: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/7.jpg)
Transient Path Loss Problem
MIT
AT&T
Avi
Patrick
Sprint
Link Down
Eventually, Patrick withdraws path from AT&T and Avi
AT&T and Avi stop sending packets to Patrick
AT&T announces the Sprint path to Patrick Traffic flows
Temporary Packet LossTemporary Packet Loss
Significant loss happens in today’s Internet, even when connected to Tier 1s
![Page 8: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/8.jpg)
How do we solve Patrick’s problem?
Tell Patrick a failover path before the link fails
rather than after it, as is often the case in current BGP
![Page 9: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/9.jpg)
AT&T advertises to Patrick “AT&T Sprint MIT” as a failover path
Link Fails Patrick immediately sends traffic on failover path
MIT
AT&T
Avi
Patrick
Sprint
Link DownNo Loss ! No Loss !
Help Patrick Help You!
![Page 10: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/10.jpg)
Our Solution: Two Simple Rules
Routing Rule: Each router advertises only one failover path and only to the next hop router on its primary path
Forwarding Rule: When routers receive packets from the next-hop interface for their primary path they forward them along the failover path
![Page 11: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/11.jpg)
Guarantee: A router is guaranteed to see no BGP-caused packet loss during convergence, if it will have a valley-free path to the destination at convergence
Guarantee: A router is guaranteed to see no BGP-caused packet loss during convergence, if it will have a valley-free path to the destination at convergence
![Page 12: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/12.jpg)
Helps Even When Deployed in a Single AS
Patrick JoeR1
R2
To MIT
Link Down
R1 offers Failover path to R2
R2 U-turns packets – back to R1
No Loss ! No Loss !
Currently Patrick drops packets even if he knows an alternate path
MIT
![Page 13: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/13.jpg)
Experimental Results
• Router-Level Simulation over the full Internet AS-graph from Routeviews and RIPE BGP Data Use inference algorithms to annotate links with customer-provider or peer relationships Add border routers based on the connections to other AS Used internal MRAI of 5s and external MRAI of 30s
• For each experiment: Random destination Take down a Random Link Find the duration of packet loss for routers using the down link which have a path after convergence Run for 1000 Randomly Chosen Links and Destinations
![Page 14: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/14.jpg)
BGP
0
0.2
0.4
0.6
0.8
1
0 10 20 30 40 50 60
BGP
Fra
cti
on
of
Rou
ters
Duration of Packet Loss in Seconds
~20% See 30s or More of Packet Loss
Significant Benefit Running Only in AT&T
![Page 15: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/15.jpg)
0
0.2
0.4
0.6
0.8
1
0 10 20 30 40 50 60
BGP
BGP - MRAI = 0
Fra
cti
on
of
Rou
ters
Duration of Packet Loss in Seconds
Significant Benefit Running Only in AT&T
Setting MRAI to 0 still leaves Significant Packet Loss
Twice the Number of Updates for Both AT&T and Customers
![Page 16: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/16.jpg)
0
0.2
0.4
0.6
0.8
1
0 10 20 30 40 50 60
BGP
BGP - MRAI = 0
Only AT&T Deploys
Fra
cti
on
of
Rou
ters
Duration of Packet Loss in Seconds
Significant Benefit Running Only in AT&T
Less than 3% if only AT&T adopts
![Page 17: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/17.jpg)
0
0.2
0.4
0.6
0.8
1
0 10 20 30 40 50 60
BGP
Deploy Everywhere
Fra
cti
on
of
Rou
ters
Duration of Packet Loss in Seconds
Running Everywhere Eliminates All Packet Loss
Full Benefit Once Running Everywhere
![Page 18: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/18.jpg)
Little Additional Overhead
0
100
200
300
400
BGP Our ApproachThousa
nds
of
Update
s
312K
325K
Less than 5% more updates network wide
![Page 19: Eliminating Packet Loss Caused by BGP Convergence Nate Kushman Srikanth Kandula, Dina Katabi, and Bruce Maggs](https://reader035.vdocuments.site/reader035/viewer/2022062518/5697bf711a28abf838c7e03f/html5/thumbnails/19.jpg)
Conclusion
• Offer a failover path only to next-hop neighbor
• Eliminates packet loss resulting from BGP convergence
• An adopting ISP reduces loss even if no other ISP cooperates
• Simple Mechanism
• Solves Problem
• Deployable