The FPC CPU might go high during BGP convergence. The route convergence will take place and it will take FPC CPU to install the routes in FIB.
The FPC CPU might go high during route convergence and may remain high until the routes gets converged fully even after router convergence completed by RPD and PFE, SRRD may take time to converge the routes.
The SRRD takes FPC CPU until the routes fully converged by the daemon. The FPC CPU will go high for 5-10 minutes depending on the type of routes, the scale of routes, and the complexity of the next hop types.
Below is the test conducted in which total 1.07M IPv4 and 800K routes were installed repeatedly by clearing the bgp neighborship. It is observed that even after routes were installed in the RIB and FIB, a significant CPU utilization of FPC observed.
Router> show bgp summary
Apr 02 22:59:50
Groups: 1 Peers: 1 Down peers: 0
Table Tot Paths Act Paths Suppressed History Damp State Pending
inet.0
358311 358311 0 0 0 115995
inet6.0
0 0 0 0 0 0
Peer AS InPkt OutPkt OutQ Flaps Last Up/Dwn State|#Active/Received/Accepted/Damped...
10.61.63.21 500 56311 5 0 6 59 Establ
inet.0: 358311/358311/358311/0
inet6.0: 0/0/0/0
Apr 02 23:01:33
1070422 1070422 0 0 0 0
800001 800001 0 0 0 252917
10.61.63.21 500 60639 8 0 6 2:42 Establ
inet.0: 1070422/1070422/1070422/0
inet6.0: 800001/800001/800001/0
Router> show chassis fpc
Apr 02 23:01:35
Temp CPU Utilization (%) CPU Utilization (%) Memory Utilization (%)
Slot State (C) Total Interrupt 1min 5min 15min DRAM (MB) Heap Buffer
0 Empty
1 Online 42 100 5 98 72 54 2048 39 22
2 Online 44 100 4 98 66 44 2048 37 22
Apr 02 23:03:48
1 Online 42 100 6 99 83 60 2048 44 22
2 Online 45 100 1 99 78 51 2048 46 22
Router> start shell command "cprod -A fpc2 -c \"show route all summary\"" | no-more
Apr 02 23:04:12
Protocol Route Tables:
Protocol Tables Routes
------------- -------- --------
IPv4 7 1070493---- 1.07M IPv4 Routes.
MPLS 2 8
IPv6 3 800038 -----800KIPv6 Routes
CLNP 2 3
DHCP-Snooping 1 1
Apr 02 23:04:15
1 Online 42 100 6 99 84 62 2048 46 22 ------------ FPC CPU remained high even after routes converged in RIB.
2 Online 45 100 1 99 80 53 2048 49 22
Router> start shell command "cprod -A fpc2 -c \"show sample-rr summary\"" | no-more
Apr 02 23:06:23
IPv4 Routes : 1070438
IPv6 Routes : 753644 ----- FPC CPU observed to be high as routes are still getting converged in Sample RR
- This is an expected behaviour that is when one does the "clear bgp neighbor".
- During the route convergence process, we noticed that the FPC utilization remained high.
- Although the forwarding table had converged with all the IPv4 and IPv6 routes, the FPC CPU still remained high.
- Upon further investigation, we discovered that SRRD was still in the process of converging the routes.-SRRD is an abreviation for Sampling Route Record Daemon, which is needed for jflow to work.
- Once the routes were converged in SRRD, the FPC utilization returned to normal.
- SRRD, or Sampling Route Record Daemon, uses a low-priority thread for convergence. Once the route convergence is completed by RPD and PFE, SRRD takes the FPC CPU cycles to converge.
- Therefore, the high FPC CPU usage was observed until SRRD was converged.- This is non-impacting and once a high-priority job pitches in, it will release the CPU cycles it holds.
IPv6 Routes : 753644
Apr 02 23:06:30
IPv6 Routes : 800015. ------> routes fully converged
Apr 02 23:06:38
1 Online 42 100 8 99 90 67 2048 54 22
2 Online 45 83 1 97 87 59 2048 58 22 ------ FPC 2 utilization dropped
Apr 02 23:06:42
1 Online 42 100 8 99 90 67 2048 54 22 ----FPC 1 utilisation high
2 Online 45 19 1 84 85 59 2048 58 22
Router> start shell command "cprod -A fpc1 -c \"show sample-rr summary\"" | no-more
IPv6 Routes : 535787 ---- Routes for FPC1 still getting converged
Apr 02 23:07:22
1 Online 42 100 7 99 91 68 2048 56 22
2 Online 45 20 1 51 76 57 2048 58 22
Apr 02 23:08:04
IPv6 Routes : 800015 ----- once the routes got converged in sampler RR
Apr 02 23:08:12
1 Online 42 39 8 89 90 69 2048 58 22 ------ FPC utilization came to normal
2 Online 45 20 1 32 68 55 2048 58 22