Per-Entity Metadata Working Group

[Nick Roy <>]

On 9/14/16 2:40 PM, Patrick Radtke wrote:
> On Wed, Sep 14, 2016 at 12:16 PM, Nick Roy 
> <>
>  wrote:
> > That is exactly why having at least one CDN, and preferably more than one,
> > is relevant to a failover strategy.  Geographic and logical redundancy.
> > Publishing to more than one CDN than one means you can configure clients to
> > fail over between more than one service.  Keeping the naming separate and
> > putting the failover between CDNs in the client lets clients specify
> > parameters they are comfortable with for things like latency.
>
> I was curious about the tradeoff between having clients configure CDN
> failover vs using round robin DNS that returned CNAMEs to each CDN.
> For the latter, in the event of a single CDN outage, a less
> intelligent client would at least hit a good CDN half the time and
> cache the result - but this assumes a unintelligent client will cycle
> through the DNS records returned. According to this
> http://stackoverflow.com/a/7204522/54396 Java does not do so by
> default. I'm not that familiar with .Net, but it seems to default to
> 'No' as well: 
> https://msdn.microsoft.com/en-us/library/system.net.servicepointmanager.enablednsroundrobin(v=vs.110).aspx
>
> Anyhow, my conclusion from reading is that most clients will likely
> just hit the first IP in the DNS cache and not cycle through the
> others.
>
> -Patrick
There is at least one example of a use case for multiple load balanced 
CDNs (realtime serving of Javascript client libraries for use in web 
apps), and their solution, although vaguely described, seems to have 
onerous infrastructure requirements:

https://www.jsdelivr.com/features/multi-cdn-load-balancing

Nick