On 16/03/14 22:24, Tom Herbert wrote:
> On Sun, Mar 16, 2014 at 2:22 PM, Thomas D Nadeau
> <[email protected]> wrote:
>> that's kind of what you and russ were saying (and I agreeing): iteration, 
>> evolution and refinement have often proven better than revolution.
>>
> Iteration, evolution, and refinement are key and in fact this is
> precisely a core rationale in GUE and geneve. We need the ability to
> scale and adapt the data center to changing needs and threats, this is
> the requirement. This translates into a requirement for extensible
> protocols that we can modify and evolve as needed and in short
> turnaround time.

I am sorry to disappoint you, network protocols are not sold in Walmart
in the "disposables" isle.

Standard network protocol's lifetime is > 10 years. Even if we discard
hardware out of the equation, it is a standard it is still >  5 years.

This is a result of economic realities - depreciation periods on
equipment, RFI, RFQs, costs of training staff, developing debugging,
benchmarking and analysis tools.

People adopt standards so that they do not need to reinvest into these
every 6 months. This is why it is called a "standard" and this is what
IETF does as an organization.

>
> The example that I give in security: what if next week the latest
> Snowden revelations expose some security hole in our data center-- if
> this hole can be addressed by adding a token to my packets, then I
> want to guarantee I retain the capability to do this. It's not
> acceptable that I'd have to go back to standards to change this, and
> neither is swapping out new hardware just because I extended a
> protocol.

It is a bit difficult to take seriously an example that extends RFC 3514
from one to 128 evil bit as a security measure. Especially against the
class of adversaries you are referring to.

>
> I agree that modifying existing protocols would be the best direction
> but that is going to require a real proposal which we can measure
> against (no one has suggested a reasonable alternative). In lieu of
> modifying an existing protocol, developing a a new one based on known
> techniques and mechanisms is prudent.

That is fine. Provided that you do requirements first.

I have yet to see requirements. Sorry - "This extra 128 RFC 3514 bits
will defend us against the [insert agency name]" is not a requirement
(unless it is the first of April of course).

When I see something that has a decent statement of requirements that
does not read like marchitecture we can have a proper engineering
conversation including how many standard engineering input units
(espressos) does it take to implement it.

A.

>
> Tom
>
>

_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to