As a follow-up to this, I've discovered that the CPU utilisation for
'netstack' process jumps to 100% for ~5 mins. High ping times are
observed to the L3 connected addresses (over which IS-IS runs). This
is almost certainly responsible for the loss of IS-IS adjacencies.

Cheers,

Matt

On 24 January 2011 13:23, Matthew Melbourne <[email protected]> wrote:
> Thanks Lincoln.
>
> I've manually added the CoPP entries for IS-IS/FabricPath to the
> policy, but we are still seeing IS-IS adjacencies drop, so I'm not
> convinced CoPP is the issue here. There are no other errors in the log
> at the time of the adjacency loss to point towards a cause.
>
> Very occasionally we're seeing:
>
> %MODULE-4-MOD_WARNING: Module 2 (serial: JAF1419BGPK) reported warning
> due to EOBC heartbeat failure in device 10 (device error 0xc000904d)
>
> However, these alerts are not co-incident with the loss of IS-IS
> adjacencies (though these alerts are seen for all modules), and the
> helpful advice so far is to "reseat all the modules to clear the ASIC
> buffers".
>
> Cheers,
>
> Matt
>
> On 16 January 2011 23:28, Lincoln Dale <[email protected]> wrote:
>> On 17/01/2011, at 7:02 AM, Matthew Melbourne wrote:
>>
>>> We are currently seeing IS-IS adjacencies flap on one of our pair of
>>> N7k boxes (eachN7k is dual-attached to two upstream edge routers):
>>> [..]
>>> I am wondering whether the default CoPP policy is classifying IS-IS
>>> CLNS traffic its class-default class and causing random instability
>>> (which is also visible on our ICMP monitoring):
>>>
>>> class-map class-default (match-any)
>>>      police cir 100 kbps , bc 310 ms
>>>      module 1 :
>>>        conformed 62024128193 bytes; action: transmit
>>>        violated 470896229 bytes; action: drop
>>
>> depending on when you applied the CoPP policy to the system (e.g. if this 
>> switch has been deployed for a while), it may well be that the CoPP policy 
>> wasn't specific enough to have ISIS explicitly mapped to its own class.
>>
>> over time we've added more & more classes into CoPP to make it more 
>> granular, but unless you've run the 'setup' script you won't pick those 
>> changes up with newer releases, as it would be bad practice for us to be 
>> changing users' configurations on ISSU.
>>
>>> Is there any way of specifiying IS-IS traffic within a CoPP class, in
>>> order to prevent it being policed in any way?
>>
>> ISIS uses a well know set of mac-addresses (0180.c200.0014/15) which you can 
>> use with a mac ACL for CoPP.
>>
>> most recent NX-OS 5.1 has this as the default CoPP policy:
>>
>>        mac access-list copp-system-acl-mac-fabricpath-isis
>>          10 permit any 0180.c200.0015 0000.0000.0000
>>          20 permit any 0180.c200.0014 0000.0000.0000
>>        !
>>        class-map type control-plane match-any copp-system-class-critical
>>          [..]
>>          match access-group name copp-system-acl-mac-fabricpath-isis
>>        !
>>        policy-map type control-plane copp-system-policy
>>          [..]
>>          class copp-system-class-critical
>>            set cos 7
>>            police cir 39600 kbps bc 250 ms conform transmit violate drop
>>          [..]
>>
>> if you wish to see the most recent iteration of the default policy, its in 
>> the documentation.
>> <http://www.cisco.com/en/US/docs/switches/datacenter/sw/5_x/nx-os/security/configuration/guide/Cisco_Nexus_7000_NX-OS_Security_Configuration_Guide__Release_5.x_chapter24.html#con_1072128>
>>
>>
>> cheers,
>>
>> lincoln.
>
>
>
> --
> Matthew Melbourne
>



-- 
Matthew Melbourne

_______________________________________________
cisco-nsp mailing list  [email protected]
https://puck.nether.net/mailman/listinfo/cisco-nsp
archive at http://puck.nether.net/pipermail/cisco-nsp/

Reply via email to