One of the transition mechanism for IPv6 is 6to4. This is an automatic mechanism which encapsulate IPv6 packets inside IPv4 packets. The packets are sent to relays which are located around the world. A list of these public relays are available at:

http://www.bgpmon.net/6to4.php?week=4

How these relays work?

They use the same technology as for the DNS: anycast. The IPv4 relay is always at 192.88.99.1 on 192.88.99.0/24 network, and this network is advertised via BGP to the rest of the world. It means your 6to4 client will look for the closest (in terms of BGP network) relay. All the relays advertise the same route, therefore your closest router may see a few routes to the various relays in the world and will choose the closest, the same way your client choose the closest DNS root server.

So your 6to4 device will encapsulate an IPv6 packet via IPv4 to the closest relay. This relay will convert it to IPv6 send it and get the answer back. The IPv6 address is well defined and is part of the 2002::/16 network. Similarly the relay will advertise this route on the BGPv6 network, so the packet can come back this way. Or via the closest on IPv6 network 6to4 relay. Communications can be quite asymmetric with 6to4.

Format of an 6to4 IPv6 address
2002:IPv4 address of your 6to4 client:00:MAC-48 derived address of your IPv6 device
16bits:32bits:16bits:64bits=128bits

If you are an ISP get a 6to4 relay in place

It seems a lot of devices offer 6to4 out of the box, and sometimes unknown to the user. Because the list of public relays are limited, it is important that ISPs, looking at offering IPv6, set their own relay. The relay does not need to be public, BGP allows to advertise routes to some networks only. So an ISP could make their relay known only to their customers. This would improve user experience, and also allows the ISP to control the relay. 6to4 is not a bad method, but MTU configuration could be quite a challenge when you don't fully control the network.