Per-Entity Metadata Working Group

[Nick Roy <>]

On 9/14/16 12:42 PM, Cantor, Scott wrote:
> On 9/14/16, 2:27 PM, 
> "
>  on behalf of Tom Scavo" 
> <
>  on behalf of 
> >
>  wrote:
>
> > For per-entity metadata, it's a totally different story. Fail over
> >     needs to be automatic, and the various servers need to be
> >     strategically located in geographic regions such that latency is
> >     minimized (or at least acceptable).
> >
> >     Does that help?
> It does, but I had the impression CDNs were part of the failover story here, that IC 
> would probably operate the "backbone" and perhaps a redundant site that might 
> be manual or automatic (preferably the latter of course) but that some kind of CDN 
> strategy would be the core of the model.
>
> I really thought the whole idea with CDNs was that the client doesn't need to 
> know. I fetch from mdq.incommon.org and the Internet's magical wonderfulness 
> routes me to the best option and constantly monitors for options being down 
> without me having to care.
>
> I'm starting to come to grips with GSLB here in a campus context, and that's 
> what they (will) do for us. Mercifully, this is a stateless service unlike an 
> IdP, so the issues are much simpler.

GSLB is notoriously fraught.  Allowing clients to fetch from multiple 
upstream services is potentially a good hedge if less than 3 nines 
worries you.

Nick

> -- Scott