Re: [Detnet-dp-dt] an attempt on math for global pools of labels

[Loa Andersson <[e-mail]>]

Yes,  it is true that it is rough, but normally I want a scaling factor
of at least 10 (or more).

/Loa

On 2017-02-22 19:41, Carlos Jesús Bernardos Cano wrote:
> Hi Loa,
>
> I agree it is worth doing this kind of rough estimate. I guess we need
> also some insights from use cases to do a better estimate.
>
> On Wed, 2017-02-22 at 18:59 +0800, Loa Andersson wrote:
>> Folks,
>>
>> Figures are not scientific, just off the top of my head.
>>
>> Assume that we have 400 edge nodes and the receiving and sending
>> capability of each node is 2400 G and the average DetNet flow is 1G,
>> and the traffic matrix is close to even.
>>
>> This means that each node will need 2400 Labels to manage as a pool
>> for
>> the d-pw labels if we go with the idea to make them unique for the
>> entire domain. 400 * 2400 = 960 000 which brings us close to
>> exhausting
>> the label space.
>
> So you assume that all edge nodes are saturated, right? Is this a
> reasonable assumption? I have some doubts, but this is just a personal
> feeling.
>
> And, what do we assume of the inner topology/nodes capacities? because
> unless the topology is really meshed, that number of flows would
> generate an aggregated throughput (especially considering that some
> nodes will be doing replication of packets) in some nodes that is
> probably higher than what the links can support, no?
>
> Thanks,
>
> Carlos
>
>>
>> I real life I don't think that the average speed is as high as 1G, if
>> it
>> were "only" 500 M, the label space would be exhausted.
>>
>> /Loa

-- 


Loa Andersson                        email: [e-mail]
Senior MPLS Expert                          [e-mail]
Huawei Technologies (consultant)     phone: [phone]