Re: [BEHAVE] Home NAPT44 - How many ports?

["Reinaldo Penno (repenno)" <[e-mail]>]

There are some interesting measurements on this "background TCP
radiation", i.e., how much state (and bandwidth) a home consumes even when
there is no active use.

On 6/5/13 3:57 PM, "Rajiv Asati (rajiva)" <[e-mail]> wrote:

>> >If one is doing CGN, wouldn't it be reasonable to point customers' at a
>> >recursive DNS server with an interface inside the CGN?
>> 
>> Yes. That's what I suggest. But some people use, say, Google's
>> DNS/OpenDns/etc and in some other cases the network is not setup
>> correctly.
>
>And in some cases, it is not possible depending on where the NAT function
>is placed and where the DNS server is placed. I recently ran into this in
>a large mobile network design.
>
>Nonetheless, it is desired, but it is not really a big deal, since UDP
>NAT usage tends to be a lot less than that TCP NAT usage (barring few
>exceptions).
>
>> >This seems to also suggest that having a *caching* recursive DNS(SEC,
>> >HOMENET+, mDNS+) server inside the customer router is also a big win.
>> 
>> Yes, it is.
>
>Well, DNS resolver with or without proxy is a big win, I would say.
>
>
>Cheers,
>Rajiv
>
>
>> -----Original Message-----
>> From: Reinaldo Penno (repenno)
>> Sent: Wednesday, June 05, 2013 12:28 PM
>> To: Michael Richardson; [e-mail]
>> Cc: Poscic, Kristian (Kristian); Rajiv Asati (rajiva); Softwires-wg list
>> ([e-mail]); [e-mail]; Erik Kline ([e-mail])
>> Subject: Re: [BEHAVE] Home NAPT44 - How many ports?
>> 
>> 
>> 
>> On 6/5/13 1:23 PM, "Michael Richardson" <[e-mail]> wrote:
>> 
>> >
>> >>>>>> "repenno" == repenno  <Reinaldo> writes:
>> >    repenno> On the other hand, as Rajiv captured,the number of
>> >    repenno> UDP sessions can be much larger than the number of
>> >    repenno> TCP. Because the way
>> >    repenno> dynamic webpages are constructed today, there are
>> sometimes
>> >    repenno> literally 100s
>> >    repenno> of DNS requests to download a single page.
>> >
>> >If one is doing CGN, wouldn't it be reasonable to point customers' at a
>> >recursive DNS server with an interface inside the CGN?
>> 
>> Yes. That's what I suggest. But some people use, say, Google's
>> DNS/OpenDns/etc and in some other cases the network is not setup
>> correctly.
>> 
>> >
>> >This seems to also suggest that having a *caching* recursive DNS(SEC,
>> >HOMENET+, mDNS+) server inside the customer router is also a big win.
>> 
>> Yes, it is.
>> 
>> >
>> >--
>> >]               Never tell me the odds!                 | ipv6 mesh
>> >networks [
>> >]   Michael Richardson, Sandelman Software Works        | network
>> >architect  [
>> >]     [e-mail]  http://www.sandelman.ca/        |   ruby on
>>rails
>> >   [
>> >
>