Hi Weiguo,

You still missing it .. sorry. There is nothing centralized in my model.

As I said my model does not exist today - it is new proposal - so any
attempts to map it to existing RFC will not be successful.

I said on the picture that NVE is still there on the hosts (compute
nodes) in the role of PE1.

VRFs on PE2/ASBR are only for aggregation purposes.

Moreover I thought this was obvious but solution I proposed is a
superset .. it means ASBR for some VPNs can be just pure vanilla
option B and for those which have a lot of prefixes can be enhanced
with per VRF aggregation function based on the vrf select mode per
VN_ID.

Effectively what I am proposing in this long thread is to define few
modes of VN_ID allocations (just like we have few modes of VPN label
allocation on a classic PE).

For some VPNs you may have VN-ID directly mapped to VPN label and for
some where aggregation of VPN routes makes sense you map VN_ID to a
local VRF.

Thx,
R.

On Wed, Jul 9, 2014 at 9:27 AM, Haoweiguo <[email protected]> wrote:
> Hi Robert,
> Your illustration is very clear. Your model is centralized NVO3 gateway + 
> Inter-as Option-B, our model in the draft is distributed NVO3 gateway + 
> Inter-as Option-B.
> Your model is very good and easy to be deployed, it will be a great 
> complementary to this solution.
> In your model PE2 have all VRFs of the data center. For the traffic data  
> from NVO3 side to MPLS VPN side, VRF is selected based on VN ID. For the 
> traffic data from MPLS VPN side to NVO3 side, VRF is selected based on MPLS 
> VPN Label instead of sub-interface between ASBRs.
> In our model , gateway is located on each NVE instead of on the PE2/ASBR, VRF 
> is located on each NVE, only VN ID and MPLS VPN Label switching table exists 
> on the ASBR, VRF don't need to be setup on the ASBR.It is more similar to 
> vanilla inter-as Option-B.
> Thanks
> weiguo
> ________________________________________
> 发件人: [email protected] [[email protected]] 代表 Robert Raszuk 
> [[email protected]]
> 发送时间: 2014年7月8日 20:56
> 收件人: Haoweiguo
> 抄送: [email protected]
> 主题: Re: 答复: 答复: 答复: Solicit reviews and comments on 
> draft-hao-l3vpn-inter-nvo3-vpn-00
>
> Hi Weiguo,
>
>> In this email, your model are clearly described
>
> I don't think so based on your reply. Your interpretation is
> completely unrelated to my model :) Sorry.
>
> Let me try to illustrate:
>
>                               Data Center
>                         WAN
>
>  VMs/namespaces
> hypervisor + kernel (PE1)
>    compute node
>                 |
>                 |
>                  --------------------- PE2/ASBR --- option B --- ASBR
> ------------- PE3
>
>
> I am describing PE2/ASBR network element where you combine plain
> option B with the VRF select based on VN_ID PE.
>
> DC can be operated by any provider so can WAN. Those are completely
> not related to each other except agreement to establish L3VPN option B
> peering session between corresponding ASBRs.
>
> PE2 is acting as WAN side aggregator on a per VRF basis. There is no
> option A in this model anywhere.
>
> L3 VPN architecture does not provide such model hence perhaps you were
> trying to map it to something which does not really exist in any RFC
> document and this was the src of your confusion.
>
> I suggest if you want to build scalable interconnect between DC based
> virtualized resources and arbitrary WAN you use this model. Of course
> for completenss such mode must also provide independent access from
> Internet either plain or over IPSec, but I assume this is obvious and
> here we are only focusing on hybrid L3 VPN interconnect case.
>
> PE1 may be just a data plane or it could be also control plane and
> data plane. The choice there is flexible.
>
> Hopefully with the picture this is more clear now.
>
> Regards,
> R.
>
>
>
> On Tue, Jul 8, 2014 at 12:11 PM, Haoweiguo <[email protected]> wrote:
>>
>> Hi Robert,
>> Thanks for your clear description.
>>
>> In this email, your model are clearly described, it's different from the 
>> model in our draft. In your model, NVO3 gateway and PE are integrated in a 
>> single device. In our model, NVO3 gateway and PE are two independent devices.
>> If data center network and MPLS VPN WAN network belongs to two different 
>> companies (for example, data center network belongs to Microsoft, MPLS VPN 
>> network belongs to AT&T), NVO3 gateway and PE function normally are splitted 
>> into two devices instead of integrating into single device. In this case, 
>> inter-as solution between NVO3 gateway and PE should be used, there are two 
>> solutions of Option-A and Option-B. In this draft, Option-B solution is 
>> described in detail.
>>
>> Pls see my other replies inline with [weiguo].
>>
>> Then the draft is wrong as Option-A does not use any MPLS encapsulation.
>> [weiguo]. Sorry, it's a slip of the pen. Yes, Option-A has no MPLS 
>> encapsulation, native IP forwarding is used between ASBRs. Every VPN people 
>> should know the procedures.
>> The correct description is:
>> This is a inter-as option between NVO3 and MPLS VPN network. When traffic 
>> from internal NVO3 network is forwarded to external MPLS VPN network, the 
>> ASBR(NVO3 Gateway)
>> terminates into a VRF and performs IP lookup,  then forwards the traffic to 
>> peer ASBR. This mechanism is called Option-A in the draft.
>>
>> You are confusing on which side of ASBR VRFs sit. I am describing the
>> model where ASBR is an option-B ASBR with in the same time integrated
>> PE towards the DC side. There is no option A.
>> [weiguo]: I understand what's your meaning now, we are not discussing the 
>> same scenario. In your description, PE and NVO3 gateway are integrated in a 
>> single device, so no Option-A here. If MPLS VPN network and data center 
>> network belongs to same company(such as AT&T, china telecom, and etc), your 
>> model are suitable, PE and NVO3 gateway are integrated on same device, VN ID 
>> is bounded to a VRF on the device.
>> In our draft,  PE and NVO3 gateway are splitted into two devices, PE belongs 
>> to carrier network(such as china telecom, AT&T,and etc), NVO3 gateway 
>> belongs to data center(such as Facebook, Microsoft, Tencent, and etc), in 
>> this case inter-as Option-A or Option-B solution should be used. In this 
>> case, NVO3 gateway in data center is ASBR1, PE in carrier network is ASBR2. 
>> If sub-interface is used to separate VPN traffic between ASBR1 and ASBR2, 
>> the solution is Option-A. If MPLS VPN Label is used to separate VPN traffic 
>> between ASBR1 and ASBR2, it is Option-B solution, the procedures is 
>> described in detail in this draft.
>>
>> Note also for clarity that packets are arriving at such PE from DC
>> side not traditionally based on bounding to interface to a VRF but
>> based on VN_ID bounding to a VRF. Some vendors call this "vrf select"
>> mode.
>> [weiguo]: Yes, in your model, bounding interface to VRF isn't needed, only 
>> VN ID bounding to VRF exists.
>>
>> Sorry if this model/architecture was not clear in my former mail.
>> [weiguo]: I understand your model now, your model are different from the 
>> model described in our draft.
>>
>> Thanks
>> weiguo
>> ________________________________________
>> 发件人: [email protected] [[email protected]] 代表 Robert Raszuk 
>> [[email protected]]
>> 发送时间: 2014年7月8日 17:06
>> 收件人: Haoweiguo
>> 抄送: [email protected]
>> 主题: Re: 答复: 答复: Solicit reviews and comments on 
>> draft-hao-l3vpn-inter-nvo3-vpn-00
>>
>> Hi Weiguo,
>>
>>
>>> What's wrong with IP lookup per VPN on ASBR ?
>>>
>>> [weiguo]:  This is a inter-as option between NVO3 and MPLS VPN
>>> network. When traffic from internal NVO3 network is forwarded to
>>> external MPLS VPN network, the ASBR(NVO3 Gateway)
>>> terminates into a VRF and performs IP lookup, MPLS encapsulation
>>> and then forwards the traffic to MPLS VPN network. This
>>> mechanism is called Option-A in the draft.
>>
>>
>> Then the draft is wrong as Option-A does not use any MPLS encapsulation.
>>
>>
>> You are confusing on which side of ASBR VRFs sit. I am describing the
>> model where ASBR is an option-B ASBR with in the same time integrated
>> PE towards the DC side. There is no option A.
>>
>> Note also for clarity that packets are arriving at such PE from DC
>> side not traditionally based on bounding to interface to a VRF but
>> based on VN_ID bounding to a VRF. Some vendors call this "vrf select"
>> mode.
>>
>> Sorry if this model/architecture was not clear in my former mail.
>>
>>
>>> In extreme case, 16M VN/Tenants and several million end systems
>>> need to be supported in a data center,  inter-as option-A solution
>>> has following drawbacks:
>>
>> I am not talking about option A so the following list does not apply.
>> Moreover each solution needs to scale and I would not put any hard
>> limit regardless what it is.
>>
>> So if particular box reaches its capacity you add another box to act
>> as a gateway. That's it.
>>
>>
>>> On ASBR1, VN ID is allocated per MPLS VPN Label received from peer
>>> ASBR2 in MPLS VPN network.
>>
>> That is very inefficient and I would argue that architecturally wrong.
>> This puts all the WAN churn and load towards all compute nodes which
>> just does not belong there and brings no value.
>>
>>> If remote PE in MPLS VPN network allocates MPLS VPN Label per VPN,
>>
>> There is no such a thing. At best you can allocate a VPN label per PE
>> per VRF, but this would require a lot of reconfiguration of the entire
>> WAN L3VPN network which may not be welcome by number of providers.
>>
>> Best regards,
>> R.

Reply via email to