Re: [dhcwg] Anyone interested in continuing draft-ietf-dhc-dhcpv6-prefix-pool-opt?

["Leaf Yeh" <[e-mail]>]

[RM] The main difference here is that RAAN option was used to provide per-client configuration/route while if I understand correctly your proposal this option will be used to configure a per-box/per-relay route or per-group of users (per-service) route on the relay as both you and Med confirmed. 

The better understanding is that OPTION_PREFIX_POOL is used for a group of 'clients', whose client's PD prefixes is from the same pool. The OPTION_PREFIX_POOL is associated with those OPTION_IA_PDs for the clients. The DHCPv6 server has full knowledge (including lease) about this kind of resource in your address plan.


[RM]  This is my point: you repeat the same option in every Relay-reply message  in order to do a one time/per-relay configuration in my opinion this does not seem to be very efficient, but I may be wrong.

In the mechanism described in this draft, the relay includes OPTION_ORO in the Relay-Forward message to request the server to response OPTION_PREFIX_POOL included in the Relay-Reply message. The relay don't know whether the prefix will be delegated to the client is from a new prefix pool, it will include OPTION_ORO in the Relay-Forward of Solicit/Request message to request the server to response OPTION_PREFIX_POOL. After T1/T2 of this client, the relay who keep status of each lease of the client's PD prefixes in the same pool might know the 'Lease of the Pool' has not expired; but it doesn't know whether the server had changed the status of the pool according to policy on the route aggregation or not, it still will include OPTION_ORO in the Relay-Forward of Renew/Rebind message to request the server to response OPTION_PREFIX_POOL.


[RM] My point was  not about the packet length, but more mode of operation: you are asking the Relay to process multiple times the same option in order to install a route that is already there.   

The draft has not touched how to implement the aggregated route (eg. black-hole route) into the routing table yet, but to provide the PE router the information about the prefix pools and their status, then to sync the status of pools on the server with the status of pools on the relay through the above 'piggyback' option in the PD messages. The draft believe your PE router will have enough logic to keep (or handle) each lease of the client's PD prefix and the status of the associated aggregated routes (or the prefix pools). That means that your PE router might need to establish a new table to keep the status of the associated prefix pools (or the aggregated routes); only after the status of the prefix pools changed, then your PE router need to take action to add or withdraw the associated route.


Best Regards,
Leaf


---------------------------------------------------------------------------------
From: Roberta Maglione [mailto:[e-mail]] 
Sent: Monday, August 26, 2013 6:04 AM
To: Leaf Yeh
Cc: [e-mail]; [e-mail] WG; Ralph Droms
Subject: Re: [dhcwg] Anyone interested in continuing draft-ietf-dhc-dhcpv6-prefix-pool-opt?

Hello Leaf,
thanks for your comments please see inline [RM]
Best Regards
Roberta
On Sun, Aug 25, 2013 at 11:51 AM, Leaf Yeh <[e-mail]> wrote:

Not sure you have interest to read draft (RAAN, draft-ietf-dhc-dhcpv6-agentopt-delegate) again.
I believe Bernie (or Tomek) have better answer for this question.

[RM] The main difference here is that RAAN option was used to provide per-client configuration/route while if I understand correctly your proposal this option will be used to configure a per-box/per-relay route or per-group of users (per-service) route on the relay as both you and Med confirmed. 

Roberta - How would it work with this new option? How does the DHCP Server know if it has to add or not this option if it is not based on client profile or on client request?

The OPTION_PREFIX_POOL will be included in the Relay-Reply message for DHCPv6-PD Reply to the clients.
Pls. read the draft (OPTION_PREFIX_POOL) for the details, Madam.


Roberta - Do you plan to insert the option in all  the DHCP Server replies/to all clients?

Yes, almost every Relay-Reply message for Reply to the clients as described in the draft.

[RM]  This is my point: you repeat the same option in every Relay-reply message  in order to do a one time/per-relay configuration in my opinion this does not seem to be very efficient, but I may be wrong. 
 

Roberta - I assume the router is going to install the route once and what does it do if it gets the same routes more the one? Just ignore?

Read the draft, keep in mind you have a smart PE with strong CPU. :-)


Roberta - It seems like a waste of cycles on the router.

The OPTION_PREFIX_POOL proposed by the draft to include in the Relay-Reply message is not more than 22 bytes with a fixed format.
Your concern here sounds not a problem.

[RM] My point was  not about the packet length, but more mode of operation: you are asking the Relay to process multiple times the same option in order to install a route that is already there.   

Roberta - In my opinion having global parameters on DHCP Server seems  a major change in the DHCP model.

The DHCPv6 server already has the ‘global parameters’, such as the prefix pool … as you mentioned above.
The draft (OPTION_PREFIX_POOL) does intend to enhance the (deployment) capability of DHCPv6 server. Pls. read my last reply mail to you & Med's mail again.
The solution with a newly defined option is a solution we thought with a minimal protocol change.
New capability from the new protocol design do need a little change on your habit for the employment in your deployment.

 


Best Regards,
Leaf



----------------------------------------------------------------------------------------------------------------------------------------
From: [e-mail] [mailto:[e-mail]] On Behalf Of Roberta Maglione
Sent: Saturday, August 24, 2013 2:52 AM
To: [e-mail]
Cc: [e-mail] WG; Ralph Droms
Subject: Re: [dhcwg] Anyone interested in continuing draft-ietf-dhc-dhcpv6-prefix-pool-opt?

Hi Med, Leaf,
I fully understand the requirement for network automation versus manual configuration, but I still don’t think defining a new dhcp option is the right direction to go in order to do routes aggregations in a router.
As somebody else already pointed out here, if you need a way to automatically interact with the router i2rs is the right place to discuss your requirements.
In addition you said:
➢  I confirm this option is not a per-customer configuration parameter.
 If it is not a per-client configuration why  would you like to use DHCP for that?
I understand that in DHCP Server there are already some global parameters like the pool of addresses to be used for the clients, but my understanding of DHCP behavior , and please correct me if I’m wrong, is that the DHCP Server provides a per-client configuration.
How would it work with this new option? How does the DHCP Server know if it has to add or not this option if it is not based on client profile or on client request?
Do you plan to insert the option in all  the DHCP Server replies/to all clients?
I assume the router is going to install the route once and what does it do if it gets the same routes more the one? Just ignore? It seems like a waste of cycles on the router.
In my opinion having global parameters on DHCP Server seems  a major change in the DHCP model.
Thanks
Roberta

On Fri, Aug 23, 2013 at 2:15 AM, <[e-mail]> wrote:
Hi Roberta,

Yes, as indicated in the document, manual configuration is an option… but it has its limits too.

This proposal is a contribution to add more automation to network configuration without requiring an additional dynamic protocol to drive how aggregates are built in a router co-located with a requestor, and therefore interact in a more dynamic fashion with a routing protocol (e.g., drive route withdrawals, etc.).

Of course, some routers can offer some features to optimize the size of routing tables and prevent from injecting (very) specific entries. But still this behavior is implementation-specific and does not provide the same aggregation level as the one proposed in this document.

Unlike implementation-specific behaviors, this proposal is deterministic since it is fully controlled by the entity which has the full knowledge of prefix related states and network policies: e.g., the server has the knowledge of prefix assignment, prefix assignment policies, prefix aggregates, etc.

I confirm this option is not a per-customer configuration parameter.

Cheers,
Med


De : Roberta Maglione [mailto:[e-mail]]
Envoyé : jeudi 22 août 2013 22:31
À : BOUCADAIR Mohamed IMT/OLN
Cc : Ralph Droms; [e-mail] WG

Objet : Re: [dhcwg] Anyone interested in continuing draft-ietf-dhc-dhcpv6-prefix-pool-opt?

Hello,
Maybe I’m missing something here, but I’m struggling to see the value added by this new option in terms of route aggregation functionality.
Today with IPv4 if I need to summarize some routes I manually configure on the router a summary/aggregate route and I announce it into the routing protocol. Moving to IPv6 you could do the same thing, I don’t quite get what’s wrong with that?
You say you would like to have an automatic way to tell the PE to aggregate the routes, but if I understand correctly the proposal what you are doing here is only moving the configuration of the summary route from the PE to the DHCPv6 Server; what do you really save here?
In addition the route aggregation is not a per customer configuration, it would be per box or per service configuration so why do you want to add it to customers’ profile in DCHPv6 Server?
Thanks
Roberta

On Thu, Aug 22, 2013 at 9:45 AM, <[e-mail]> wrote:
Re-,

IMHO, draft-ietf-dhc-dhcpv6-agentopt-delegate does not cover the same objectives as in draft-ietf-dhc-dhcpv6-prefix-pool-opt.

draft-ietf-dhc-dhcpv6-prefix-pool-opt aims to provide a dynamic means to trigger route advertisement actions and to control the route aggregates to be injected using a routing protocol. For example, a router can be told by the DHCP server to advertise an aggregate even if not all individual prefixes are assigned to customer located behind that router. This is a measure that can help in optimizing routing tables and avoid injecting very specific routes. Snooping the assignment and then guide the route advertisement actions may not be lead to the same optimized routing tables, because there will be "holes" that will prevent aggregating routes.

Having an explicit channel like the one specified in draft-ietf-dhc-dhcpv6-prefix-pool-opt is superior IMHO.

Cheers,
Med


>-----Message d'origine-----
>De : [e-mail] [mailto:[e-mail]] De la part de
>Ralph Droms
>Envoyé : jeudi 22 août 2013 14:48
>À : Alexandru Petrescu
>Cc : [e-mail] WG
>Objet : Re: [dhcwg] Anyone interested in continuing draft-ietf-dhc-dhcpv6-
>prefix-pool-opt?
>
>
>On Aug 21, 2013, at 9:24 AM 8/21/13, Alexandru Petrescu
><[e-mail]> wrote:
>
>> Hi,
>>
>> One point I think is essential is the installment of routes in the DHCP
>> Relay upon Prefix Assignment.
>>
>> The base DHCPv6 Prefix Delegation RFC does not stipulate that DHCP must
>> install a route in the DHCP Relay upon delegation.
>>
>> This draft seems to at least assume it, and to describe much more about
>> it: how various parts of assigned prefixes are aggregated and
>communicated.
>>
>> I support it.
>
>After a quick read, draft-ietf-dhc-dhcpv6-agentopt-delegate seems to have
>been aimed at the same problem.  If I have that right, it might be
>instructive to review the dhc WG mailing list discussion that lead to the
>abandonment of draft-ietf-dhc-dhcpv6-agentopt-delegate.
>
>- Ralph
>
>>
>> Alex
>>
>> Le 21/08/2013 14:41, [e-mail] a écrit :
>>> Hi Tomek,
>>>
>>> I do still think draft-ietf-dhc-dhcpv6-prefix-pool-opt documents a
>>> useful feature in order to have more automation and also control
>>> routes aggregation instead of relying on proprietary behaviors of
>>> each implementation. Of course, part of these objectives can be
>>> achieved if routes are installed manually or use an out of band
>>> mechanism to enforce routing aggregation policies. Still, the
>>> proposal in draft-ietf-dhc-dhcpv6-prefix-pool-opt is superior
>>> because the DHCP server has the knowledge of the prefix assignments;
>>> and therefore routes can be triggered with dhcpv6 .
>>>
>>> A way to progress with this document is to target the Experimental
>>> track. Based on the experience that will be gained in real
>>> deployments, the status can be revisited if required.
>>>
>>> Cheers, Med
>>>
>>>> -----Message d'origine----- De : [e-mail]
>>>> [mailto:[e-mail]] De la part de Tomek Mrugalski
>>>> Envoyé : lundi 19 août 2013 16:52 À : dhcwg Objet : [dhcwg] Anyone
>>>> interested in continuing draft-ietf-dhc-dhcpv6- prefix-pool-opt?
>>>>
>>>> During Berlin meeting chairs asked if there is still interest in
>>>> the prefix-pool-option. There was nobody interested in the work in
>>>> the room. The unanimous consensus in the room was to drop it. I
>>>> just wanted to confirm that on the list.
>>>>
>>>> If you are interested in this work, want to support it and
>>>> participate in it, please let us know by replying to the mailing
>>>> list. Otherwise we'll drop this work and mark that draft as a dead
>>>> WG document.
>>>>
>>>> Please respond within 2 weeks (until Sep. 2nd).
>>>>
>>>> Bernie & Tomek _______________________________________________
>>>> dhcwg mailing list [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/dhcwg
>>> _______________________________________________ dhcwg mailing list
>>> [e-mail] https://www.ietf.org/mailman/listinfo/dhcwg
>>>
>>>
>>
>>
>> _______________________________________________
>> dhcwg mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/dhcwg
>
>_______________________________________________
>dhcwg mailing list
>[e-mail]
>https://www.ietf.org/mailman/listinfo/dhcwg
_______________________________________________
dhcwg mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/dhcwg