On 12/7/12 10:32 AM, "Eric Osterweil" <[email protected]> wrote:

>
>On Dec 7, 2012, at 7:56 AM, Montgomery, Douglas wrote:
>
>> I just want to point out that I was on a different train.  Discussing
>>the
>> supposition that query&cahce arch's behave differently that batch pull
>> archs in either startup or steady state performance.  I was not
>>commenting
>> on the whole trust/business model discussion.
>> 
>> 
>> I was replying to Eric's suggestion that DNS-like systems are an example
>> of the on-demand architectural choice.
>
>Doug, this is getting silly.  If you have a system that _requires_ full
>state of all data producers in the entire world before it can begin, then
>you are talking about a _much_ different architecture than one that only
>queries for _relevant_ information when it becomes relevant.  If you use
>the latter for the former's goals, or the former for the latter's goals
>then so be it, but they are designed with different precepts.

What is silly is our failure to communicate.

If the initial immediate requirement of a demand driven system is to
individually query for 90%+ of the total information in the system, how is
that different that pre-fetching it all?

DFZ Router with 500K unique routes needs to validate its RIB using a
demand-based validator that just came up.

How would you like the demand system to perform those resolutions?  What
is your time estimate for how long the cold started demand driven
"validator" will take to validate all of those entries?

You are right that they are different architectural models.  No argument
there.   Their use for the problem we are discussing, will result in that
difference being moot from a performance scaling perspective.

No argument that the architectures are different.

We seem to differ on that difference is important ... And probably on
which would scale better.

dougm

_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to