Re: [PCN] Comments on HOSE edge behaviour discussion

[ken carlberg <carlberg at g11.org.uk>]

Georgios,

Personally, I would be more cautious in the wording of your document regarding EMCP.  For instance, I wouldn't assume the per-flow or per-destination is the default case in all instances.  Instead, I would the more neutral stance and state that the these along with per-packet configurations exist, and that in order to use the Hose model, the per-flow/per-packet MUST be used.

-ken


On Nov 16, 2009, at 4:36 AM, Georgios Karagiannis wrote:

> Hi Ken
> 
> Thanks for the clarification! 
> In my reasoning I have assumed that the default setting of ECMP is 
> per-flow or per-destination form.
> 
> We could include this assumption in the HOSE edge behaviour draft!
> 
> Best regards,
> Georgios
> 
> 
>> -----Original Message-----
>> From: ken carlberg [mailto:carlberg at g11.org.uk] 
>> Sent: vrijdag 13 november 2009 15:52
>> To: Steven Blake
>> Cc: Georgios Karagiannis; pcn at ietf.org
>> Subject: Re: [PCN] Comments on HOSE edge behaviour discussion
>> 
>> the answer is configuration and per-implementation dependent. 
>> A number of implementations have a default setting of 
>> per-flow or per-destination form of ECMP, which is what would 
>> be needed for RSVP.  The danger arises if an in-path router 
>> is configured for per-packet ECMP -- some 
>> vendors/implementations support this, some don't.
>> 
>> -ken
>> 
>> 
>> On Nov 12, 2009, at 7:57 PM, Steven Blake wrote:
>> 
>>> On Thu, 12 Nov 2009 16:32:58 +0100, "Georgios Karagiannis"
>>> <karagian at cs.utwente.nl> wrote:
>>> 
>>>>> Steven Blake: ECMP support - not safe to assume that path 
>> signaling
>>>> message will follow the same path as the data - so not safe to say 
>>>> that
>>> this
>>>>> scheme solves the ECMP support issue.
>>>> 
>>>> Georgios: Steve, please note that from the point of view of the 
>>>> followed communtication path for signaling messages and 
>> data packets, 
>>>> the same assumptions are used as the ones applied and used by RSVP 
>>>> PATH
>>> and
>>>> data packets. Thus if the RSVP signaling works to support 
>> QOS, then 
>>>> the
>>> proposed ECMP
>>>> solution will also work.
>>> 
>>> Georgios,
>>> 
>>> I don't believe that is correct.  Can you comment Francois?
>>> 
>>> 
>>> Regards,
>>> 
>>> // Steve
>>> _______________________________________________
>>> PCN mailing list
>>> PCN at ietf.org
>>> https://www.ietf.org/mailman/listinfo/pcn
>> 
> 
>