[c-nsp] Redundant switch fabric

[Justin C. Darby]

Mike,

Just to chime in here a bit with some experience - we've had Nexus 7K 
switch backplane modules fail - unless you are pushing near 100% 
backplane utilization you don't even notice until it emails you or your 
config monitoring program notices the failed module. In recent NX-OS 
releases, In Service Software Upgrades are working properly 100% of the 
time for us, and outside of the fact it can take 3-4 hours to upgrade a 
fully loaded switch, there's no real downtime if you've got working port 
redundancy across modules, and modules only go down one at a time like 
they're supposed to.

Considering how distributed and redundant components of the switch are - 
it's pretty unlikely you'd run into huge redundancy problems with any 
single component. I don't have enough N7K's to play with Virtual Port 
Channels (vPCs), but it'd be interesting to see if they have any issues 
when upgrading switches. vPCs can add extreme (and usable) redundancy to 
multi-chassis design, if you want to go a step farther.

Justin

P.S. Comments made here are my own and should not in any way be 
considered an endorsement by the U.S. Federal Government.

Brad Hedlund wrote:
> Mike,
> The 6500 and 4500 have the "switch fabric" on the supervisor engines, so by
> having dual supervisors, you in effect have a redundant fabric.
>
> The 6748 actually has 4 traces, each 20G.  2 traces connect to the active
> supervisor containing the active switch fabric.  The remaining 2 traces are
> standby connections to the standby supervisor/fabric.  So, when a supervisor
> engine and its fabric fails, the 2 standby traces are enabled and the full
> 40G of bandwidth remains.  You never, under normal circumstances, have only
> a single trace active on 6748.  Newer versions of IOS provide a "hot
> standby" fabric feature which allows this fabric trace switch over to happen
> faster - roughly 50ms.
>
> For the best in redundant designs, consider the Nexus 7000, where the switch
> fabric is decoupled from the supervisor engines into a series redundant
> "fabric modules" installed into the back of the switch.  Should a supervisor
> engine fail in Nexus 7000 there is ZERO impact to the switch fabric, because
> the supervisor engine does not forward data plane traffic.
>
> Cheers,
>
> Brad Hedlund
> bhedlund at cisco.com
> http://www.internetworkexpert.org
>
>
> On 3/31/09 9:05 AM, "Mike Louis" <MLouis at nwnit.com> wrote:
>
>   
>> I have a solution design that requires redundant switch fabrics. I am
>> interpreting this beyond just have redundant supervisors meaning redundant
>> backplanes on the switch cards. Do the 6500 and 4500 support redundant
>> fabrics? Will a 6748 function with one trace failed?
>> ________________________________
>> Note: This message and any attachments is intended solely for the use of the
>> individual or entity to which it is addressed and may contain information that
>> is non-public, proprietary, legally privileged, confidential, and/or exempt
>> from disclosure. If you are not the intended recipient, you are hereby
>> notified that any use, dissemination, distribution, or copying of this
>> communication is strictly prohibited. If you have received this communication
>> in error, please notify the original sender immediately by telephone or return
>> email and destroy or delete this message along with any attachments
>> immediately.
>> _______________________________________________
>> cisco-nsp mailing list  cisco-nsp at puck.nether.net
>> https://puck.nether.net/mailman/listinfo/cisco-nsp
>> archive at http://puck.nether.net/pipermail/cisco-nsp/
>>     
>
>
>
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>