On 12/07/2012 15:43, Truman Boyes wrote:
> Brian, I agree that diffserv alone would not address the condition if more 
> granular QoS/CoS is required. If we think of NVE as PE a device then I 
> believe we have the necessary components. 
> 
> Ingress classification on the VAP as well as the potential to classify on the 
> VNIF will handle mapping from header details to appropriate queue.
> 
> Marking of outer tunnel header as we do today with exp bits or gre ip header 
> will ensure end to end PHB. 
> 
> The missing link is how to handle the condition you describe; two streams, 
> same traffic class, different tunnels. Maybe the provider should provide 
> distinction of queue/class if there is an expectation of behavior 
> differences. Or we will need to use the second VNID (briefly described in the 
> framework doc) as a means to provide greater differentiation.
> 
> Building a form of call admission control  for tenant streams to the network 
> is looking for a world of hurt. IntServ can be our testament to this truth. ;)

To be clear, in the diffserv context, admission control does not imply state per
stream; it really implies a token bucket per diffserv class and per tunnel. If
you limit things to a handful of diffserv classes, that ought to scale.

Alternatively, provide too much bandwidth. That may well be cheaper.

   Brian

> Kind regards,
> Truman Boyes
> 
> 
> On Jul 12, 2012, at 3:04 AM, Brian E Carpenter <[email protected]> 
> wrote:
> 
>> Truman,
>>
>> On 12/07/2012 05:01, Truman Boyes wrote:
>>> Hi Selvam,
>>>
>>>
>>> Yes I agree this does need to be discussed. There could be a couple methods
>>> in achieving this:
>>>
>>>
>>>   - The virtual bridge on the hypervisor of the NVE can handle the
>>>   rate-limiting ingress or egress from the TES on the VAP.
>>>   - If the encapsulation is IP between the NVE, we could ensure that
>>>   appropriate DSCP/ToS markings are copied into the IP header based on
>>>   configured mappings of tenant QoS priority. This way standard physical
>>>   network devices can also appropriately provide the intended per hop
>>>   behavior.
>>>   - The IP ToS or DSCP values from packets generated from TES could be
>>>   simply copied into encapsulation IP header; if a corresponding VAP or
>>>   policy existed on the VNI.
>> Diffserv alone cannot ensure sharing between two competing streams
>> of the same class of traffic that happen to occur in two different tunnels.
>> You would definitely need admission control of some kind at *all* tunnel
>> ingresses, as far as I can see. Whether that scales is another question.
>>
>> FYI, diffserv handling in tunnels is documented in RFC 2983.
>>
>>    Brian Carpenter
>>
>>> Just some quick thoughts on this at midnight ...
>>>
>>> Kind regards,
>>> Truman
>>>
>>>
>>> On Wed, Jul 11, 2012 at 7:55 PM, Selvam Ramanathan <
>>> [email protected]> wrote:
>>>
>>>> Hi Truman, ****
>>>>
>>>> ** **
>>>>
>>>> Lets Take a case where Tenant A and Tenant B are configured. ****
>>>>
>>>> Since  Tenant A and Tenant B are using the same physical network over the
>>>> multiple overlay network , ****
>>>>
>>>> How do we make Sure Tenant A gets the bandwidth he needs & Tenant B is not
>>>> overwhelming the ****
>>>>
>>>> Network . ****
>>>>
>>>> ** **
>>>>
>>>> Let’s take a case where  Tenant A might be running Video Streaming Servers
>>>> on his VM/Endpoint & ****
>>>>
>>>> From Tenant A’s  perspective its required for the Virtual DC  to guarantee
>>>> his bandwidth requirements.  ****
>>>>
>>>> ** **
>>>>
>>>> Regards, ****
>>>>
>>>> Selvam ****
>>>>
>>>> ** **
>>>>
>>>> ** **
>>>>
>>>> ** **
>>>>
>>>> *From:* Truman Boyes [mailto:[email protected]]
>>>> *Sent:* Tuesday, July 10, 2012 7:02 PM
>>>> *To:* Selvam Ramanathan
>>>> *Cc:* [email protected]
>>>> *Subject:* Re: [nvo3] bandwidth requirements of the tenants in
>>>> draft-narten-nvo3-overlay-problem-statement-02.txt****
>>>>
>>>> ** **
>>>>
>>>> What are these issues relating to bandwidth that you are referring to that
>>>> are specific to overlays or to the lack of overlays? ****
>>>>
>>>> ** **
>>>>
>>>> Regards,****
>>>>
>>>> Truman****
>>>>
>>>> ** **
>>>>
>>>> On Tue, Jul 10, 2012 at 7:40 PM, Selvam Ramanathan <
>>>> [email protected]> wrote:****
>>>>
>>>> Hi , ****
>>>>
>>>> ****
>>>>
>>>> Should the overlay-problem-statement-02 also talk about the ****
>>>>
>>>> issues arising out of the bandwidth requirements of the tenants ? ****
>>>>
>>>> ****
>>>>
>>>> Regards, ****
>>>>
>>>> Selvam ****
> 

_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to