Re: [Spud] SPUD for Datacenters

[Caitlin Bestler <[e-mail]>]

On 3/30/15 11:24 AM, Mirja Kühlewind wrote:
> Hi Caitlin,
>
> just a quick response. To my experience designing thing for the Internet is usually harder that for data centers. However, what works in the Internet should usually also work in data centers. (If that’s the most efficient way to do things in a data center, is a different question). However, I think it is most important that we can come up with something that can be use in the Internet. And as this is today hard problem, I’d prefer to focus on this part.
>
> Mirja
>
>
I would agree that getting things to work for the Internet is harder.

What I am suggesting is that we at least think about the datacenter 
impact for each specific method we come up with.
Careful phrasing of requirements can result in no additional work for 
the Internet deployment but enable an efficient
datacenter implementation as well. But if we don't do that check before 
we freeze language we will end up forcing
some inefficiency for datacenter implementations.

Datacenter specific optimizations, such as enhanced interactions for 
short round-trip-times, should be deferred.
Make it work everywhere first.