[c-nsp] Nexus 7k and OTV

[Lincoln Dale]

On 26/07/2010, at 10:33 PM, Matthew Melbourne wrote:

> From talking to our Cisco SE, the initial release of OTV will likely
> on support a low-ish number of VLANs, with the likelihood that the
> number will increase over time as additional testing is performed.

like all new technologies introduced, we start with relatively conservative values and gradually increase the scale numbers.  OTV is no different.

our guidelines as far as scale in the initial release are as follows:
 - 3 Overlays each of 3 Sites, 2 Edge Devices per site (6 Total)
 - 128 OTV extended VLANs
 - 12000 MAC Addresses across all the extended VLANs
 - 100 Multicast Data Groups

there aren't "hard" limits, i.e. the system won't stop you going above them but my suggestion is that if you feel you need to go higher than these numbers then have your Cisco AM/SE talk to the relevant folks about your scale requirements.

> A
> back-to-back vPC between pairs of Nexus make work between two DCs, but
> separate physical links for L3 interconnects may be an issue.

note that there doesn't need to be "separate L3 physical links" across the WAN / DCI, rather its that the traffic on the OTV "overlay" is L3.


> Given
> this may scale to more than two DCs, solutions such as VPLS/A-VPLS
> begin to look attractive.

if you say so!
while VPLS / A-VPLS certainly is used by many people for data center interconnect, its not without its challenges.
> 
> BTW, is is permissible to use a vPC peer link between two Nexus for
> non-vPC traffic; thinking in the scenario where the access layer
> design comprises of squares rather than triangles.

yes, you can use the vPC peer-link for non-vPC traffic but certainly there are some network design considerations.
would it be the same VLANs?

i assume if you're creating a 'square' then the blocked link would be on the access layer since the STP root would be in the Agg layer?
not sure what benefit you'll get from a square in a vPC world....


cheers,

lincoln.


> 
> On 22 July 2010 13:13, Lincoln Dale <ltd at cisco.com> wrote:
>> On 22/07/2010, at 8:16 PM, Matthew Melbourne wrote:
>> 
>>> Is it possible to extend two vDCs between Data Centres using OTV (pair
>>> of N7k on each site) - it's not clear how OTV uses vDCs to extend the
>>> L2 domain.
>> 
>> yes, its possible.  there are 3 methods:
>> 
>> 1. OTV runs inside a Virtual Device Context.  for 2 VDCs you could run OTV in each of them.
>>   this approach requires 'WAN' connectivity to each VDC - perhaps a downside if its not already there.
>> 
>> 2. if you don't have overlapping VLANs between the VDCs then you can trunk the VLANs (L2) from one VDC to another using a crossover external cable (1G or 10G or Nx1G/10G depending on your traffic levels), then use a single OTV instance to transport all the VLANs between sites.
>> 
>> 3. you could run a single instance of OTV within each VDC (similar to #1) with the 'WAN' IP connectivity to the OTV cloud achieved via a single VDC with a crossover external cable (1G or 10G, doesn't really matter) @ L3 between VDCs to the single VDC that has WAN connectivity (similar to #2).  i.e. sort of a hybrid of the above two.
>> 
>>> Also, is it correct that Unicast-only transport support (and hence the
>>> use of an Adjacency Server) is not supported initially (looking at the
>>> BRKDCT-2049 Networkers' presentation).
>> 
>> correct.
>> 
>> 
>>> 
>>> Alternatively, are there other mechanisms to achieve L2 connectivity
>>> for two separte L2 domains over a pair of redundant links between Data
>>> Centres?
>> 
>> quite a few ways of doing it, but OTV does solve some unique problems and issues with the various other approaches.
>>