[dns-operations] prefetching and thundering herds

[Tony Finch]

Mark Delany <b9w at charlie.emu.st> wrote:
>
> But in terms of thundering herds, it's still seem no worse than if none of the stub
> resolvers cached.
>
> However, I can see how if the choice is between stub caches which "smear" TTL (regardless
> of mechanism) and stubs which don't "smear", the smearers are probably a little nicer to
> recursive resolvers which incur high costs queuing duplicate queries.

Right.

> But then I have to wonder, is there a fundamental reason why recursive
> resolvers perform non-linearly as the arrival rate of identical queries
> increases?

It's the bimodal behaviour between cache hits and misses, and the higher
the query rate the more queries arrive while the cache is being refilled.
I think it only gets nonlinear if the recursive server's query
deduplication is nonlinear.

I was probably wrong to worry about thundering herds - I was thinking
about them because I started wondering about synchronized cache expiry in
a somewhat different context. Then I noticed that stub caches might stymie
prefetching...

Tony.
-- 
f.anthony.n.finch  <dot at dotat.at>  http://dotat.at/
Viking, North Utsire: Variable 3 or less, becoming southerly 4 or 5,
occasionally 6 later in north. Slight, occasionally moderate later. Showers.
Good.