Re: [clouds] draft-yokota-cloud-service-mobility

[Vishwas Manral <[e-mail]>]

Hi Hidetoshi-san,

>> 2. Scalability issues will occur if keepalives all go to the manager
>> node. In my view there can be a heirarchy of keepalives.
>
> Are you suggesting an intermediate node that collects the keep-alives
> from some group of managed nodes and sends an aggregated message to the
> manager?
Yes I am suggesting one such approach. You have to understand the
number of resources that need to be monitored is very high.

A heirarchical approach will certainly help. Ofcourse there needs to
be redundancy at each layer so that a failure in the intermediate does
not lead to failures that may not necessarily be occuring.

>> 3. There should be a heirarchy of manager nodes too, considering the
>> number of Execution nodes that need to be managed. So there should be
>> a messaging exchange allowed between Manager and Manager node.
>
> I see. Either hierarchical structure (manager of managers) or
> peer-to-peer structure (inter-manager) will be needed when the scale
> becomes larger.
That is correct. Like I mentioned I am trying to come to a PCE model.
It is the same problem with different data. In PCE all the network
nodes and interface properties are kept for Traffic Engineering
purposes.
https://datatracker.ietf.org/wg/pce/

>> 4. All TLV and headers should have length of 16 bits atleast. 8 bits
>> is not scalable at all with the amount of information that is there.
>
> Good suggestion. Will expand the field length.
This is extremely essential and we have seen issues in older protocols
whih have used 8 bit length or type fields.

>> 5. There needs to be authentication in the packets to provide some security.
>
> Ok. Maybe, some option that can carry MAC (Message Authentication Code)
> should be added. Or, do you think the whole message should be encrypted?
> In that case, we should mandate IPSec connection between the Manager
> Node and Execution Node.
It is IPsec not IPSec. Yes we need to add a hash to the packet. If we
are working over TCP we can use the TCP-AO or TCP-MD5 for the same.

>> 6. There needs to be async messaging allowed both from server to
>> client and client to server.
> Ok. I will add something like NOTIFY manage, which is spontaneous and
> can
Great!!!

>>7. There is already a PCE framework that exists. We need to look at
>> it. It is used for simialr purposes in a TE network.
>
> Could you tell me any reference document such as RFC or conference
> paper, please?
Sent the link above.

Thanks,
Vishwas

> Regards,
> --
> Hidetoshi
>
>> Thanks,
>> Vishwas