Re: [**EXTERNAL**] RE: New Version Notification for draft-mudric-6man-lcs-01.txt

["Mudric, Dusan" <[e-mail]>]

Hi Eduard,

I agree with your proposed steps 1 & 2 for the application API. The first one is mentioned in the step 4.8. 

We will make draft updates to make the whole process clear. Other comments are inlined.

Thanks for good questions and suggestions,
Dusan.

On 2020-10-26, 4:00 AM, "Vasilenko Eduard" <[e-mail]> wrote:

>Hi Dusan,
>OK. It looks like the solution now.
>IMHO: you would better update draft to insert the logic from your last message. It is the critical piece of information to understand what you propose.

>You need 2 new API functions to call between applications and ND:
>1. to say on the client "use LLA for this GUA, if possible", then client would send NS with flag L.
>2. to say on the server " if NS has flag L in request for this GUA then respond with LLA too"

>The second API call would give the way for smooth transition: when ND stack already support this extension, but application does not support it yet.
[Dusan] Agree about the need of the second API for the smooth transition. 
              I think the ND stack should disable the return of LL till application/socket API is ready to support LL sockets or all sockets are using ANY address.

>It is evident that you would have the black hole
>if one application on the server would be attached to all addresses,
>But other application would be attached only to GUA.
>Because the 1st application would ask for GUA->LLA substitution.
>Hence, should be the possibility to disable this ND functionality on Server.
[Dusan] The solution can be to disable LL resolution on the server side till all sockets request the LL usage, or they bind to ANY address. 
               As long as there is one socket bound only to GUA, LL address resolution should be disabled. 
               Otherwise there will be a connectivity issue with that socket.

>You are asking too much: to change all hosts and to change all application to get full value.
[Dusan] I expect not too much. It is rather very common for applications to bind to ANY address.

>Albeit, it would work as soon as 3 particular instances would become upgraded:
>1. ND on particular client
>2. ND on particular server
>3. at least 1 application on this server
>If Enterprise admin do care about security - he could easily make such progress.
[Dusan] If clients open sockets on ANY address, only ND upgrade is required on the client and server sides.

>I have the temptation to ask you: why client is in the same subnet with the server? They are typically in different subnet.
[Dusan] In IP telephony, for example, there is a server on SiteA and number of phones on that site. But server is used by
              other phones on SiteB as well. Server is not isolated on one site, without phones on the same site.

Do you have any best guess for how many servers (%) have clients on the same link?
[Dusan] In IP telephony 100%. There is no site without phones.

Eduard