As we all know an IPv6 enabled host can have multiple addresses. In order to select a source address for a to-be established outbound connection, operating systems implement a source address selection mechanism that evaluates multiple source address candidates and selects the (potentially) best candidate. Criteria for this selection are defined in RFC6724 (which obsoletes RFC 3484).
To find out if there are differences as for the way various OSs implement this mechanism we performed a little study whose results can be found in this whitepaper. Those differences might be particularly relevant for data center environments or enterprise networks with a variety of heterogeneous client operating systems. If interested in IPv6 in enterprise networks this training that I’ll give in some weeks might be worth attending for some of you, too.
Enjoy reading and everybody have a great day
Enno
Hi Enno,
are you aware of a document that helps with choosing the address space (when to use PI/PA) and scope (when to use ULA/GLA or both) to use in an mid-sized (~1000 devices) enterprise network?
We’re such a typical mid-sized enterprise and are wondering what might be the best choice. Most of the books about IPv6 I’ve read are primarily covering the technical aspects but are lacking practical guidance on such topics.
I once thought that a combination of ULA (to be independent from the provider on the internal network) and PA-GLA (for Internet connectivity if needed) might work. However, since having multiple addresses on a device introduces more complexity and it seems to have become easier to acquire and use PI space (i.e. without running BGP) I now think that just using PI-GLA might be the better choice in most cases.
If possible we would like to omit both the complexity introduced by using multiple addresses/scopes and the dependency of PA-space. Actually, like the way it currently just works with running private IPv4 space and NAT if needed.
Any input on this is highly appreciated.
Thanks,
Michael