Re: [v6ops] Operational Headache: Provisioning domains

[Fred Baker <[e-mail]>]

> On Mar 30, 2019, at 11:56 PM, Brian E Carpenter <[e-mail]> wrote:
> 
> By the way, there was this a few years ago: https://tools.ietf.org/html/draft-jiang-v6ops-semantic-prefix I'm still worried by using bits in a GUA prefix for essentially semantic purposes. I fully realise we have a large supply of IPv6 prefixes. But mixing addresses and service semantics just seems very likely to cause problems sooner or later. I'm sorry, that reads like FUD, but that's my feeling.

No hats

That, BTW, was the reason I pushed back on Jiang's draft, and I have the same fundamental question about Terastream's use case. I understand both concepts to essentially say something about what address one uses for an application; putting it into the address has the characteristic of ensuring that the same set of bits is present in the packet end to end, which differs from DiffServ. It basically allows DT to make an end to end service, and change the IPv6 head to support it, with actually having to change the IPv6 header. It affects, or can affect, routing, and as I understand it required DT to allocate a truly large prefix. I find myself wondering - what if this were general? Everybody was allocating /24 or larger prefixes to get the equivalent of a DSCP to be usable end to end? The fact of asking the question seems crazy, but at some point, that seems to me to limit the availability of IPv6 address space. Is that a rathole we want the network to go down?

Just me worrying.
--------------------------------------------------------------------------------
Victorious warriors win first and then go to war,
Defeated warriors go to war first and then seek to win.
     Sun Tzu