1. John Scudder jgs@cisco.com October 24, 2000
BGP Update
John Scudder
October 24, 2000

2. Overview New stuff Tweaks, frobs, cleanup Graceful Restart
Cooperative Route Filtering, Prefix-ORF
Extended Communities
Route Refresh
Tweaks, frobs, cleanup
Revisions of base spec, Route Reflection, Confederations, Capabilities, MP-BGP

3. Base Spec draft-ietf-idr-bgp4-10.txt Fix ambiguities about MED
Actually, this part is not so new
Otherwise, mostly editorial

4. Graceful Restart draft-ramachandra-bgp-restart-03.txt a.k.a. NSF
When BGP restarts, continue forwarding for some (short) period of time
Avoid route flap
Non-intrusive upgrades, less harmful failures
Some risk of transient blackholes/loops
But route flaps have their own risks

5. Graceful Restart Mechanism
Capability includes
flags (currently restarting or not, preserved FIB or not)
timer (time to wait for me to restart)
“End of RIB” message

6. Graceful Restart Mechanism
When BGP stops, neighbors don’t drop TCP session right away, wait for timer instead
When BGP restarts, neighbors don’t flush old routes immediately — wait to converge
Old routes are flushed after convergence (or after a timer expires)

7. Graceful Restart Example
… and advertises its new RIB, replacing the marked routes.
Once B and C have sent all their routes, A runs best path, populates its FIB
BGP comes back up and establishes new sessions with B and C.
BGP halts on A. B and C’s RIBs and A’s FIB all mark routes for later deletion.
B and C send their routes to A. A stores them in its RIB.
When A has finished advertising its RIB, any marked routes which weren’t replaced are flushed.
Router A
Router B
RIB:
(routes from A)
(other routes)
RIB:
(routes from A)
(other routes)
RIB:
(new routes from A)
(other routes)
RIB:
(routes from B)
(routes from C)
RIB:
RIB:
(routes from B)
RIB:
(routes from B)
(routes from C)
Router C
FIB:
(new routes from B)
(new routes from C)
FIB:
(routes from B)
(routes from C)
FIB:
(routes from B)
(routes from C)
RIB:
(routes from A)
(other routes)
RIB:
(routes from A)
(other routes)
RIB:
(new routes from A)
(other routes)

8. Route Refresh
RFC 2918
Allows a router to request neighbor to re-send its whole Adj-RIB-Out
Permits soft reconfig without storing filtered Adj-RIB-In
Benefits: save memory
Drawbacks: extra communication, CPU when inbound policy is changed

9. Cooperative Route Filtering
draft-chen-bgp-route-filter-01.txt, draft-chen-bgp-prefix-orf-00.txt
Lets router export its filtering policy to neighbor
Community and prefix policies are specified so far
Reduces communication and CPU on both ends

10. Cooperative Route Filtering Mechanism
Outbound Route Filters (ORFs) are sent along with Route Refresh
So far communities and prefix lists are specified, simple encoding
Each AFI/SAFI can have its own ORFs
Can change filters by sending new Route Refresh request
Peer can use ORFs to filter outbound routes

11. Refresh + ORF Example ORF Comm 3 ORF Comm 2 Router A Router B
OPEN, Refresh, ORF = Comm 2
Pfx
10.0.1
10.0.2
10.0.3
10.0.4
Comm 1 2 3
Announce ,
Refresh, ORF = Comm 3
w/d , , Announce

12. Extended Communities draft-ramachandra-bgp-ext-communities-04.txt
8-byte, more structured communities
2 byte type, 6 byte value. Type determines format of value.
Value typically includes originator’s IP address or AS number
Defined types: route target, route origin, link bandwidth
Currently used for network-based VPNs

13. Route Reflection RFC 2796 Changes vs. RFC 1966 Editorial cleanup
Deployment section with some points regarding MED, avoidance of loops
No fundamental changes

14. Confederations draft-ietf-idr-bgp-confed-rfc1965bis-01.txt
Changes vs. RFC 1965
Corrections to reflect reality — particularly reverse code points for CONFED_SEQ and CONFED_SET.
Editorial changes
Expanded deployment section with discussion of MED, routing loops
No fundamental changes

15. Deployment Guidelines
RR and clusters scale by hiding routes
This changes some BGP assumptions
To avoid trouble:
Avoid overlapping clusters/sub-ASes
Set IGP metrics to prefer intra-cluster (or sub-AS) paths

16. Why Avoid Overlapping Clusters?
RR 1 A B
RR 2
Well-known problem (I hope!)
Avoid simply by making clusters/sub-ASes follow topology

17. Why Set IGP Metrics To Prefer Intra-Cluster Paths?
A, IGP 5 * * *
C, IGP 10, MED 1 * * * *
B, IGP 4, MED 2
A, IGP 6 *
C, IGP 10, MED 1
C, IGP 11, MED 1
B, IGP 5, MED 2 A B C
AS X
AS Y
MED 2
MED 1
RR 1
RR 2 5 4 1 10

18. Why Set IGP Metrics To Prefer Intra-Cluster Paths?
A, IGP 5 *
C, IGP 10, MED 1 *
B, IGP 4, MED 2
A, IGP 17
C, IGP 10, MED 1 12
RR 1
RR 2 4 5 10 A B C
AS X
AS Y
MED 2
AS Y
MED 1

19. Mailing List IETF IDR Working Group mailing list — idr@merit.edu For:
Discussion of BGP protocol itself
Discussion of operational needs, problems
Not:
“how do I build my network”
“vendor foo feature bar”