1. draft-haas-bfd-large-packets
Jeff Haas
Albert Fu
IETF Montreal 18th July 2018

2. Motivation
MTU issue may occur without any indication of link/protocol issue, as protocol hello/keepalive packets are small
In redundant topology, troubleshooting MTU issue is time consuming due to multiple ECMP paths
e.g. Traffic between two end points may have more than 16 ECMP paths in typical Core/Distribution/Access design
Current network typically involves multiple routing protocols (e.g. OSPF, ISIS, BGP, MPLS etc.)
Only ISIS has automatic MTU detection mechanism
Routing protocol hellos are handled by control plane, hence unable to use sub-second timer for fast detection
An automatic and fast mechanism for detection of MTU issue is highly desirable in high performance network.

3. Why use BFD for MTU detection?
Routing protocols (e.g. OSPF/ISIS/BGP) on modern high performance routers all support using “light-weight” BFD for faster failure detection
BFD function may be supported on distributed hardware and independent of control plane
Enable fast sub-second failure detection
Best practice routing protocol design leverages BFD
BFD failure detection (due to connectivity or MTU) will bring client routing protocols down, providing fast automatic traffic diversion
Alarm for BFD failure will enable immediate problem detection and follow up

4. bfd.PaddedPduSize Current Proposal Sample User Cases Max IP Payload
IP Hdr
UDP Hdr
BFD ip
len x
udp
src
port
udp
dst
3784
udp
len 32
chksum
layer2
bfd
ver
bfd
len 24 My
Disc
Your
Disc
Min TX
Min RX
Min
Echo 52 8 24
udp len=32
ip len=52
Proposal
IP Hdr
UDP Hdr
BFD ip
len x
udp
src
port
udp
dst
3784
udp
len 32
chksum
layer2
bfd
ver
bfd
len 24 My
Disc
Your
Disc
Min TX
Min RX
Min
Echo
Padded with 0x00
52+p 8 24
p=bfd.PaddedPduSize
udp len=32
ip len=52+ bfd.PaddedPduSize
Sample User Cases
Max IP Payload
Bfd.PaddedPduSize (p)
Internet Core
1,500
1,448 (1,500-52)
MPLS VPN Core (up to 3 labels)
1,460 (1,512-52)

5. Questions?