[c-nsp] Brief CPU spikes on 6500 Sup 720

[Aaron Riemer]

Guys,

We are still seeing these floods. They occur when our backups run and it
doesn't matter if the traffic is routed into the vlan or switched within the
same vlan as the same problem occurs.

Therefore: 
  1) asymmetric routing cannot be the problem for the L2 switched traffic 
  2) No TCN's are occurring within the VLAN STP (PVST).
  3) Packet captures confirmed that all ports on VLAN1 on one core were
receiving the unicast traffic destined for a host on this vlan.
  4) From the packet captures we were able to deduce source / dest MAC
addresses confirming that the destination MAC was in fact not aging out and
could be seen clear as day in the MAC table.

Had TAC on the line troubleshooting it for a couple of hours tonight and
still no closer to solving the problem. Mac-age time out is set to the same
as the arp timeout (4 hours)

I don't believe the issue is linked to the CPU spikes but rather the output
drops that we are seeing oversubscribing our 6548 line cards.

Anyone have any further ideas?

Thanks,

Aaron.

-----Original Message-----
From: Phil Mayers [mailto:p.mayers at imperial.ac.uk] 
Sent: Wednesday, 14 July 2010 11:14 PM
To: Aaron Riemer
Cc: 'Matthew Huff'; 'JC Cockburn'; cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] Brief CPU spikes on 6500 Sup 720

On 14/07/10 15:51, Aaron Riemer wrote:
> Yes i have read all about unicast flooding:
>
> Can occur by:
>
> 1) Asymmetric routing
> 2) Spanning Tree TCN
> 3) MAC aging out
>
> I cannot see any TCN's or Asymmetric routing so i think we may have to
> adjust the mac aging as you suggested.

If you're running HSRP, the standby node has a route for the subnet as 
well as the active e.g. traffic might flow out through the active, and 
back through the standby as follows:

host
  |
  active-standby
  |        |
(cloud)   ^
  |        |
  router --/
  |
target

from   host->active->router-target
return target->router->standby->host

...if "standby" has an ARP entry (default 4 hours) but not MAC table 
entry (default 5 minutes) it will unknown-unicast-flood the return 
traffic. If it's a lot of traffic, that will burn a lot of bandwidth...

Whether this will happen will depend on your routing topology.

>
> I am just trying to work out why the hell this has only just started
> occurring!

Well, without knowing the source of the traffic it's impossible to tell.