In a lab scenario, says between 2 routers, there are hundreds of sub-interfaces and a BGP session with BFD on each subinterface. Then BFD flaps is seen. The scale information is as followed:
- 256 subinterfaces
- 256 ebgp session enabled with BFD
- BFD timers are 50ms x 3
Addressing Type Up Init Down AdminDown
-------------------- -------------------- ------------- ----------- ----------- ---------
All All 239 [0] 8 [0] 9 [0] 0 [0]
IPv4 All 239 [0] 8 [0] 9 [0] 0 [0]
single hop All 239 [0] 8 [0] 9 [0] 0 [0]
normal 239 [0] 8 [0] 9 [0] 0 [0]
From the above output, you can see 17 of 256 sessions are not up. And HW BFD is enabled (by default in Eos), but no helps.
ghb289#show bfd hardware utilization
Chip Name Number Of HW Sessions* Maximum Number Of HW Sessions*
--------------- ---------------------------- ------------------------------
Jericho0 0 200
Jericho1 128 200
Jericho2 128 200
"show cpu counter queue" indicates high # of drop of CoppSystemBfd class, which means the receiving BFD packets exceeds the b/w limit of Copp.
ghb289#sh cpu counters queue |nz |grep -i bfd
CoppSystemBfd Et16/2 5584344 390904080 0 0
CoppSystemBfd Et51/2 783529888 57981211712 3985376 294917824
CoppSystemBfd Et51/2 783529888 57981211712 3985376 294917824
After increasing the shape/bandwidth of copp-system-bfd, there is bfd flaps anymore.
Class-map: copp-system-bfd (match-any)
shape : 25000 kbps
bandwidth : 2500 kbps
No comments:
Post a Comment