Re: [v6ops] About Req for Comments - "Transition to IPv6"

[Alexandre Petrescu <[e-mail]>]

Le 16/03/2020 à 22:20, Owen DeLong a écrit :
>
>
>> On Mar 16, 2020, at 13:40 , Alexandre Petrescu 
>> <[e-mail] <mailto:[e-mail]>> 
>> wrote:
>>
>>
>> Le 16/03/2020 à 20:42, Owen DeLong a écrit :
>>>
>>>
>>>> On Mar 16, 2020, at 04:03 , Alexandre Petrescu 
>>>> <[e-mail] 
>>>> <mailto:[e-mail]>> wrote:
>>>>
>>>>
>>>> Le 16/03/2020 à 08:05, Owen DeLong a écrit :
>>>>>
>>>>>
>>>>>> On Mar 15, 2020, at 16:32 , Matthew Petach <[e-mail] 
>>>>>> <mailto:[e-mail]>> wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Fri, Mar 6, 2020 at 11:54 AM Owen DeLong <[e-mail] 
>>>>>> <mailto:[e-mail]>> wrote:
>>>>>>
>>>>>>     > On Mar 5, 2020, at 8:51 AM, Alexandre Petrescu
>>>>>>     <[e-mail]
>>>>>>     <mailto:[e-mail]>> wrote:
>>>>>>     > Is facebook a network?  I thought of it like a server farm.
>>>>>>
>>>>>>     How does one run a distributed server farm throughout the
>>>>>>     world without connecting it with a network? Am I missing
>>>>>>     something?
>>>>>>
>>>>>>
>>>>>> I believe for a very long time, that would have accurately 
>>>>>> described Akamai.
>>>>>
>>>>> Akamai didn’t run a backbone. I’d argue that they did run many 
>>>>> rather sizable networks.
>>>>>
>>>>>> Until their announcement, long after they had established 
>>>>>> themselves in the marketplace, that they were going to deploy a 
>>>>>> network: 
>>>>>> https://pc.nanog.org/static/published/meetings/NANOG71/1532/20171003_Kaufmann_Lightning_Talk_Akamai_v1.pdf
>>>>>
>>>>> Oh, I’m very familiar with that particular project… It was in it’s 
>>>>> startup hey day during my tenure in Mr. Kaufmann’s group. That was 
>>>>> Akamai building a backbone to connect many (not nearly all) of 
>>>>> their networks together.
>>>>>
>>>>> I stand by my original statement. One cannot (usefully) run a 
>>>>> server farm without a network. Network != backbone.
>>>>
>>>>
>>>> I agree a server farm needs a network.  But that network might be a 
>>>> switched Ethernet with VLANs in a large building, or it might be 
>>>> made of long haul links (ATM, FDDI, multi-Gigabit Ethernet fiber).
>>>>
>>>> In the first case (call it VLAN) it is highly possible that VLAN in 
>>>> a building, or a set of close by buildings, are IPv6 only, and no IPv4.
>>>>
>>>
>>> This is just silly… For the server farm to be useful, that “VLAN” 
>>> still needs a larger network in front of it that eventually makes 
>>> contact (ideally in multiple diverse ways) with a larger internet.
>>>>
>>>> But if the facebook system is a geographically distributed server 
>>>> farm with a large network, it might be that at some point the data 
>>>> is carried on IPv4 even though it is IPv6 over IPv4.
>>>>
>>> In talking to Facebook’s engineers, they claim to do v4<->6 
>>> translation at the edge where necessary to talk to IPv4-only clients 
>>> and are IPv6-only throughout their datacenters and other networks.
>>
>>
>> that v4<->v6 means facebook network is not IPv6-only
>>
>
> No, it doesn’t. A few translation gateways at the very edge of the 
> network are precisely what allows the network itself to be IPv6-only.


Here' s where we differ in the definition of 'IPv6-only'.  I disagree 
with the above statement.  An IPv6-only network has no translatio 
gateways.  But it's just a definition, no problem :-)

Alex

>
> Owen
>