Using IPv6 with prefixes shorter than a /64 on a LAN

[pelle at hemmop.com (Per Carlson)]

> At AMSIX, a Cisco 12000 running IOS will get into trouble with the 170pps of
> ND seen there. AMSIX doesn't do MLD snooping so everybody gets everything
> and on IOS 12000 ND is punted to RP and when it's busy with calculating BGP,
> it'll start dropping BGP sessions.

Really? I've tried to duplicate the results in our lab, but I can't
provoke any problems at those numbers. Is it the "other" multicast
traffic that's interfering with ND?

When pounding the CPU with ~30 times more (5000pps) Neighbour
solicitations and flapping 1000 BGP IPv4 prefixes (out of 51000) every
5 seconds, I get the following load (worst case):

12k#sh proc cpu | ex 0.00
CPU utilization for five seconds: 99%/13%; one minute: 83%; five minutes: 76%
 PID Runtime(ms)   Invoked      uSecs   5Sec   1Min   5Min TTY Process
  29       19472   7944653          2  0.31%  0.07%  0.05%   0 PowerMgr Main
 160        5120      3415       1499  0.47%  0.18%  0.06%   0 Exec
 181       17016  76522129          0  0.07%  0.14%  0.15%   0 CEF RP IPC Backg
 185     1992892  19727573        101 17.91% 19.36% 20.02%   0 IPv6 Input
 213      256008   9155905         27  3.03%  2.80%  2.83%   0 BGP Router
 216     3606044    677600       5321 64.31% 45.74% 37.41%   0 BGP Scanner
12k#

Even though the load is high, there is no flaps, neither in ISIS, LDP,
BFD (3 sessions with 3 x 50 ms asynch mode) nor BGP.

When BGP Scanner is not running, the numbers are much lower:

martin#sh proc cpu | ex 0.00
CPU utilization for five seconds: 45%/16%; one minute: 82%; five minutes: 76%
 PID Runtime(ms)   Invoked      uSecs   5Sec   1Min   5Min TTY Process
 160        5192      3454       1503  0.79%  0.20%  0.08%   0 Exec
 181       17068  76522593          0  0.15%  0.15%  0.15%   0 CEF RP IPC Backg
 185     2000528  19764701        101 24.79% 19.70% 20.01%   0 IPv6 Input
 213      256976   9156110         28  3.03%  2.82%  2.83%   0 BGP Router
martin#

The hardware in question is a PRP-1 running SY9b, and the same LC
(SIP-601/SPA-5x1GE-v2) is used for both ND and BGP.

Note: When doing 10000pps ND, the LDP-adjacency with a neighbour on
the same LC did flap occasionally.

-- 
Pelle

RFC1925, truth 11:
?Every old idea will be proposed again with a different name and
?a different presentation, regardless of whether it works.