Re: [v6ops] cpe-slaac-renum: Proposed text for prefix lifetimes

[Fernando Gont <[e-mail]>]

On 8/4/20 11:57, Bernie Volz (volz) wrote:
[...]
> 1. One reason to cap the lifetime(s) in the CPE for RAs is that while the SP could adjust the lifetimes, that has a direct impact on the amount of DHCPv6 traffic as reducing the preferred lifetime will cause the renewals to occur more frequently. That is one reason why SPs do not use short preferred lifetimes. (Some SPs have many millions of CPEs.)

Agreed. Maybe one might add a sentence about this?



> 2. I do wonder whether leaving the valid lifetime alone is better than playing with both lifetimes. For most applications that don't require long lived addresses, it achieves the desired effect. For those that want long lived addresses, it allows for that (if case of outages longer than the capped valid lifetime) -- and if the application decides to try for a new address, it will use one in a new prefix if one was advertised.

Reality is that, once you have an outage of over 15', all your TCP 
sessions will be aborted, anyway. So having long valid lifetimes will 
not buy you anything in that respect.

In that sense, you'd be well enough with a Valid Lifetime of 2 * Router 
Lifetime.

The reason why we left the Valid Lifetime to a rather long value (1 day) 
is mainly for the case where you want to use such prefix for local 
communications.

However, strictly speaking, you should be using link-locals for that 
(but I realize there are issues associated with the fact that they 
require an interface to be specified along with the address) or, better, 
you should be using ULAs.

I guess, if one wanted a finer-grained recommendation, one might unfold 
the recommendation for GUAs vs. ULAs. With GUAs, you want more timely 
information. With ULAs, you don't really care that much, because it's 
local addressing (isolated from your ISP).


Going back to your orignal comment: if you don't do garbage collection 
(i.e. reasonable Valid Lifetime) on your prefixes/addresses, then that 
means you wouldn't be able to communicate with the new "owner"/user of 
the prefix -- because you consider that prefix to be local/on-link.
I'm not saying that I think there's a lot of chances that you might 
need/want to communicate with the new user of the prefix. But if for 
some reason (plain bad luck? :-) ) you needed, then you couldn't.

(that's why you don't want one-month valid lifetimes for PIOs)

Thanks,
-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492