Re: A common problem with SLAAC in "renumbering" scenarios

[Mikael Abrahamsson <[e-mail]>]

On Thu, 31 Jan 2019, Sander Steffann wrote:

> A flash cell is at least capable of 1000 to 3000 erase/write cycles. 
> Even if the prefix would change every single day that gives a lifetime 
> of 3 to 9 years. And prefixes shouldn't change that often anyway. When 
> looking at more realistic estimates line a change on average once a week 
> you'l get to 20 years. Seems more than enough for low end extra cheap 
> devices. I don't see this driving cost up, except for spending a few 
> hours on properly engineering this.

There are places where they flap your WAN (PPPoE session for instance) 
every day to renumber you. I have personally experienced this.

So it might be possible to do what you're saying but only writing partial 
information of the prefix, so skip writing lifetimes, and just keep the 
actual prefix. This is not what a typical DHCP server/client does today, 
it re-writes the file every time the lease is renewed so it knows about 
it. Some kind of compromise might be where a "hint" file is written and 
the only time this is re-written is when the prefix is changed.

OpenWrt today stores most volatile things in /tmp which is just /tmpfs. 
This has the added disadvantage that after a reboot the LAN lease file is 
empty.

-- 
Mikael Abrahamsson    email: [e-mail]