I've finally convinced $DAYJOB to deploy IPv6. Justification for the IP space is easy, however the truth is that a /64 is more than we need in all locations. However the last I heard was that you can't effectively announce anything smaller than a /48. Is this still true?
Is this likely to change in the immediate future, or do I need to ask for a /44?
A /64 is for a single link (broadcast domain, though with IPv6 multicast
domain is more appropriate).
A /48 (or /56 for end-users for some of the RIRs) is for a single
end-site ("a different administrative domain and/or a different physical
location").
If you thus have 5 end-sites, you should have room for 5 /48s and thus a
/47 is what you can justify.
If you though are not able to do transit / routing between those sites
as they are not connected one might want to get separate PI /48s for
them. But likely if you are in that camp, just asking for address space,
that you can use stably for a long time, from your network provider who
provides you connectivity is a better way to go.
Greets,
Jeroen
First:
But likely if you are in that camp, just asking for address space,
that you can use stably for a long time, from your network provider who
provides you connectivity is a better way to go.
Um, sorry I figured by the fact that I was posting on Nanog the context was clear, but I've forgotten how Nanog is now a go-to source for home network too 
The context was for what Nanog was originally intended for: We are provider-independent and peering around the world.
A /64 is for a single link …(snip)... A /48 (or /56 for end-users for some of the RIRs) is for a single end-site
Sorry, I wasn't looking for the breakdown of expected usage. I know those maps. What I was asking was whether you can PI-route a /56 or anything less than a /48 today. It's "nice" to have a few dozen of the entire Internet for each site, but totally unnecessary.
If you thus have 5 end-sites, you should have room for 5 /48s and thus a
/47 is what you can justify.
Really? One bit can flip that many ways? 
I assume you mean /45, and apparently ARIN's recommended size is /44 anyway.
I've finally convinced $DAYJOB to deploy IPv6. Justification for
the IP space is easy, however the truth is that a /64 is more
than we need in all locations. However the last I heard was that
you can't effectively announce anything smaller than a /48.
Is this still true?
Hi Jo,
The short answer to your question is:
/48 is the longest prefix from a direct RIR assignment that everyone
currently accepts via BGP.
/32 is the longest prefix from an ISP allocation that everyone
currently accepts via BGP.
As with IPv4 /24's, some folks accept longer prefixes. Not everyone.
Is this likely to change in the immediate future, or do I need to ask for a /44?
You need to ask for a /44.
Regards,
Bill Herrin
one of the downsides to v6 is the huge amnt of space the folks expect you to announce.
lots of space to do nefarious things. that said. if you select your peers carefully and don't mind
a bit of hand crafting, you can /96 and even /112
that said, get a /32 and assign/announce /48s...
/bill
Couple of errors there, Jeroen…
1. 5 /48s is at least a /45, not a /47 which is only 2 /48s.
2. Joe lives in the ARIN region where allocations and assignments are
done on nibble boundaries, so his /45 would be rounded up to
a /44 (as would a /47) anyway.
Owen
Jo Rhett wrote:
I've finally convinced $DAYJOB to deploy IPv6. Justification for the
IP space is easy, however the truth is that a /64 is more than we need
in all locations. However the last I heard was that you can't
effectively announce anything smaller than a /48. Is this still true?Is this likely to change in the immediate future, or do I need to ask
for a /44?
/48 is the new /24
randy
/48 is the new /24
Except you can stuff pretty much into one. I'm numbering my entire
workplace from one. 1500 people and 26 offices. Our v4 is a constrained
/16, which is enough. But not more.
RIPE Labs had an interesting article about filtering of /48 prefixes earlier this year that might be of some interest to you: https://labs.ripe.net/Members/emileaben/ripe-atlas-a-case-study-of-ipv6-48-filtering
There's also a useful RIPE Labs article on general prefix filtering lengths from August last year: https://labs.ripe.net/Members/dbayer/visibility-of-prefix-lengths
Edward Dore
Freethought Internet