Re: [addr-select-dt] Proposed default policy table

[Mohacsi Janos <[e-mail]>]

On Mon, 12 Jul 2010, Aleksi Suhonen wrote:

> On 07/12/10 07:38, Arifumi Matsumoto wrote:
>>> I have one other topic to add which I can't remember whether I posted 
>>> about recently or not, so apologies if I'm repeating myself.
>>> 
>>> In undertaking some work recently on router-based hints for address 
>>> selection, we found that we could improve address selection behaviour for 
>>> hosts by using a new ICMP-based message to query a router or route server 
>>> for an ordered list of address pairs to use for given candidate addresses. 
>>> The issue that came up was that if the host were to cache the result, and 
>>> we chose to cache it within the RFC3484 policy tables, can we 
>>> change/augment the current syntax/format to accommodate that?
>> 
>> That approach also should have an issue of policy merge/conflict.
>> 
>> When we think about merging process, we may have to think about this
>> kind of approach, if we think it looks like some advantages in merging.
>
> The Linux Way of recent days seems to be to replace /etc/<software>.conf with 
> /etc/<software>.d/*.conf so in Linux specifically you could implement it like 
> this:
>
> /etc/gai.conf:
> 1. default rules that can be overriden
> 2. include instruction for /etc/gai.d/*.conf
> 3. system wide rules that can't be overriden
>
> And then your address selection policy update daemon(s) would update their 
> own file(s) in /etc/gai.d/ safely.
>
> I don't have a good understanding on how other operating systems have 
> implemented RFC3484 so I can't comment how universal this would be.

I think it much more complicated:
- You have default ruleset initialised by the operating system.
- You can  have host specific ruleset configured by the particular node 
configuration files - usually loaded by the startup process like 
from /etc/<software>.conf (or /etc/<software>.d/*.conf) into the kernel

- You might opt from loading something into the kernel from the previous 
cache. I don't think it is reliable. OS should send out something like 
confirmation, whether it is still valid or ICMPv6/DHCP policy table 
solicitation/request.

  - You can have a daemon process, that actually initate this policy table 
solicitation/request and process the incoming policy table and insert the 
ruleset into the kernel.


I don't believe that overriding a particular rule either from 
configuration file or any  policy table distibution method would be 
possible. Every operating system or local setup migth be different 
therefore complete table override is possible, but modifying one rule 
not....

Best Regards,
 		Janos Mohacsi


-


>
> --
> 	Aleksi Suhonen
> 	Department of Communications Engineering
> 	Tampere University of Technology
> _______________________________________________
> addr-select-dt mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/addr-select-dt
>