Re: IPV6 Anycast for streaming

[Christopher Morrow <morrowc.lists () gmail com>]

On Sun, Aug 12, 2012 at 5:25 AM, Oliver
<olipro () 8 c 9 b 0 7 4 0 1 0 0 2 ip6 arpa> wrote:
> On Sunday 12 August 2012 04:11:25 Voice of the Blind ™ Network Operation
> wrote:
> > Hello,
> > is a anycasted Prefix a good idea for Streaming?
> > here's what we're thinking about:
> > 1. get a /48 from a LIR or a RIR and anounce it through Hurricanne
> > through several Tunnel server including Singapor, New york and
> > Amsterdam
> > 2. run 3 diferent Icecast server  in each of the locations where
> > frankfurd would be the Master Server while the 2 others would be a
> > slave relay
> > 3. Assign same Ipv6 address to each of the icecast server where we
> > would be using other prefix to do Inter server communication
> > so if someone is in North America would go through new york, if is in
> > south east asia would be throug SG, otherwise africa/Europ would go
> > through Frankfurd
> > is that a good idea for streaming performance ?
> > any other suggestion is welcome
> > Thank you
>
> What you are talking about isn't really Anycast from the perspective of the
> rest of the world since what you're saying sounds like you'll just be single-
> homed behind HE.

anycast can take many forms, including anycast inside a single ASN
which to the outside world looks like a single announcement (or even
an announcement covered by an aggregate) - 4.2.2.2 falls into that
sort of deployment, I believe.

> Regardless of whether you peer with HE in multiple locations or just one, the
> AS path is still going to be the same and will enter into 6939 at whatever the
> BGP-closest point is - the Anycast aspect essentially exists only between you
> and HE.

well, if he's proposing a 'single server' (simplistic on my part) at
each of 3 locations which then has some routing gear (or quagga?)
linking it to HE ... that's 'anycast' (in the same way as many root
server operators do it)

> Providing HE are configuring a discriminator across their iBGP routers to take
> the shortest path at entry points you're not directly peered to, the whole
> thing should Just Work, sudden topology changes notwithstanding.

agreed. One would probably be interested in not the 'anycast' so much
as 'why are you doing 3 deployments'? and maybe: "is redundancy and
traffic direction something better done in DNS than BGP"? questions.

-chris