Per-Entity Metadata Working Group

[Thomas Lenggenhager <>]

Hi Scott,

Thanks for your detailed answer.

We use IP Anycast since many years with our two geographically distant 
WAYF instances. It does not share the load evenly but the failover works 
great. Only once we had an issue due to some misconfiguration within the 
IP Anycast 'mechanism' when a failed instance was not properly removed.

Since a few years we use the IP Anycast also for our IdPs (the one for 
SWITCH as well all the hosted/managed ones we operate). We use an 
active-standby setup with a manual hand-over process.

So we will reuse IP Anycast for our per-entity metadata service we will 
liekly setup next year.

Thomas

On 21.10.16 14:07, Scott Koranda wrote:
> Hi Thomas,
>
> > I guess CDNs in combination with IP Anycast could be a suitable approach.
> > The single IP Anycast address should avoid the issues with for Java and
> > libcurl Scott brought up.
>
> Scott Cantor asked Brent Putman to look in detail at the Java
> situation. You can see his detailed reply at
>
> http://marc.info/?l=shibboleth-dev&m=147639642710536&w=2
>
> In short it appears that Java "does the right thing".
>
> We also have "hints" that libcurl "does the right thing". See
> for example this code in GitHub:
>
> https://github.com/curl/curl/blob/master/lib/connect.c#L1156
>
> More research and testing with libcurl would be helpful if
> anyone has time.
>
> Having written all that...
>
> > A quick Google search provided the following three CDNs that offer IP
> > Anycast:
> >   https://blog.cloudflare.com/a-brief-anycast-primer/
> >   https://www.maxcdn.com/blog/anycast-ip-routing-used-maxcdn/
> >   https://www.keycdn.com/support/anycast/
> >
> > What do you think about this approach?
>
> I think anycast CDNs should also be considered.
>
> LIGO has some limited experience with anycasting. We didn't
> investigate CDNs using anycasting but we did set up a test/dev
> tier of IdPs with anycasting to test an approach to "global
> HA".
>
> It looked like it would work well for us but then the people
> with the knowledge left the project so the approach has been
> put on hold for the time being.
>
> Thanks,
>
> Scott K for LIGO

--
SWITCH
------
Thomas Lenggenhager
P.O. Box, 8021 Zurich, Switzerland
phone +41 44 268 1505  direct +41 44 268 1541
https://www.switch.ch