[Nfd-dev] [NFD - Task #2010] RIB cost guidelines

[Dave Oran (oran)]

On Oct 16, 2014, at 5:32 PM, Lan Wang (lanwang) <lanwang at memphis.edu> wrote:

> 
> 
> On Oct 16, 2014, at 2:20 PM, Dave Oran (oran) <oran at cisco.com> wrote:
> 
>> 
>> On Oct 16, 2014, at 3:05 PM, Syed Obaid Amin <obaidasyed at gmail.com> wrote:
>> 
>>> The similar problem in IP is solved by Administrative Distance (AD), I think we can use the same approach here? 
>>> 
>> “Solved” is somewhat of an overstatement. Perhaps a gross overstatement.
> 
> What problems does AD have in your experience?
> 
1) It’s too coarse
2) It doesn’t help with traffic engineering
3) It interacts in strange ways with LocalPref and MultiExitDisc in BGP

Basically, it’s a minimal mechanism that people can use when they don’t really care to “solve” the route optimization and selection problem.

I wasn’t claiming it didn’t “work” in the sense it caused loops or led to route oscillation. I claim is doesn’t “solve” any of the important global route optimization and selection issues we care about.

NDN is an opportunity to rethink a lot of these questions in interconnecting routing at boundaries, because it does not need to be beholden to choices that were made 25 years ago with different technical, political, and economic tradeoffs.

For example, given the amount of state leakage NDN requires via stateful forwarding, does it still make sense to try to hide path characteristics across peering boundaries the way BGP does?

Given the extra computing we have to do anyway for hierarchical name-based forwarding, is it now such big deal compared to the rest of the overhead to do metric mapping and rescaling at boundaries, especially if we can learn from feedback provided in forwarding? 

Are our loop detection and suppression capabilities in NDN forwarding good enough that we can relax the total ordering that falls out of giving precedence to administrative distances?

Can we generalize at least somewhat from admin distance to “admin boundary cost increment”.

I’m not necessarily advocating any or all of these ideas above, only encouraging us to not say “admin distance worked for IP, therefore we can treat this as a 'solved problem' and stop thinking”.

DaveO.


> Lan
>> 
>>> Obaid
>>> 
>>> On Thu, Oct 16, 2014 at 12:54 PM, Lan Wang (lanwang) <lanwang at memphis.edu> wrote:
>>> I don't have a good answer to this question.  I suggest that we discuss this at the next nfd call.
>>> 
>>> Lan
>>> 
>>> 
>>> Begin forwarded message:
>>> 
>>>> From: <redmine-noreply at named-data.net>
>>>> Subject: [NFD - Task #2010] RIB cost guidelines
>>>> Date: October 7, 2014 11:49:34 AM CDT
>>>> To: Undisclosed recipients:;
>>>> 
>>>> Issue #2010 has been updated by Junxiao Shi.
>>>> Routes can come from several sources:
>>>> 
>>>> 	• static route (manually configured by nfdc register)
>>>> 	• intra-domain routing protocol
>>>> 	• inter-domain routing protocol
>>>> 	• local application
>>>> 	• remote prefix registration (laptop registers a prefix on the access router toward the laptop)
>>>> The costs from all those sources are not comparable, but forwarding is using an order derived from those costs.
>>>> 
>>>> Task #2010: RIB cost guidelines
>>>> 	• Author: Junxiao Shi
>>>> 	• Status: New
>>>> 	• Priority: Normal
>>>> 	• Assignee:
>>>> 	• Category: Docs
>>>> 	• Target version:
>>>> Provide guidelines for how routing protocols should assign RIB cost.
>>>> 
>>>> Each routing protocol (including static) has its own way to assign costs. Costs from different routing protocols are not necessarily comparable.
>>>> Forwarding does not interpret or use Cost assigned by routing protocol, but has access to an ordered list of nexthops in the FIB entry. This order is determined by the Costs of Routes, even if they could come from different routing protocols.
>>>> 
>>>> This Task is to provide guidelines on how routing protocols should assign RIB cost, so that Costs from different routing protocols can be comparable.
>>>> Some routing protocols may use non-integer costs; the guidelines should recommend how to convert such costs into integers.
>>>> 
>>>> You have received this notification because you have either subscribed to it, or are involved in it. To change your notification preferences, please click here: http://redmine.named-data.net/my/account
>>> 
>>> 
>>> _______________________________________________
>>> Nfd-dev mailing list
>>> Nfd-dev at lists.cs.ucla.edu
>>> http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev
>>> 
>>> 
>>> _______________________________________________
>>> Nfd-dev mailing list
>>> Nfd-dev at lists.cs.ucla.edu
>>> http://www.lists.cs.ucla.edu/mailman/listinfo/nfd-dev
>> 
>