Limiting memory usage

[Kevin Darcy]

phn at icke-reklam.ipsec.nu wrote:

> bert hubert <ahu at ds9a.nl> wrote:
> > In article <aa9f60$b09 at pub3.rc.vix.com>, phn at icke-reklam.ipsec.nu wrote:
> >>
> >> bert hubert <ahu at ds9a.nl> wrote:
> >>> In article <aa5n2v$1hj at pub3.rc.vix.com>, phn at icke-reklam.ipsec.nu wrote:
> >>
> >>>> I don't think you can (in a reasonable way) limit a nameservers
> >>>> memory hunger. It's a function of the clients usage and TTL values
> >>>> set in the RR.  Trying to enforce lower TTL will break the
> >>>> original RR-publishers suggestion.
> >>
> >>> DJBDNS dnscache does this just fine. You tell it how much memory it should
> >>> use and it uses exactly that, dropping old entries to store new ones.
> >>
> >>
> >> A nameserver is _supposed_ to cache data up to the time given
> >> in the original TTL. By silently truncationg the cache it breaks
> >> the function it's supposed to do.
>
> > It will answer your question just fine. It has to go out on the net to do
> > so. So you are claiming that only a nameserver with infinite memory is rfc
> > compliant.. boy.
>
> The DNS design builds upon caching as a vehicle to speed up accesses.
> Time to cache is set by the publisher of data. Now this may lead to
> a conflict where a nameserver is "too busy" for it's avaliable memory/TTL.
> This is an admin error. Any nameserver that is correct configured
> will meat a balance where old entries times out in teh same pace as
> new ones is stored.
>
> Trying to limit memory usage is trying to fix the problem in the wrong end.
> Aggree that one has to, in order to keep services up in the short run.
> The solution is to BMM.
>
> There is no _demands_ that any nameserver always must be able to keep
> it's cache intact, shure. But a nameserver that randomly drops whatever
> it feels for is no good solution.

Most pre-expiration algorithms drop the oldest or most-soon-to-expire entries.
This is hardly "random". If the nameserver pre-expires an entry that would have
expired "naturally" in 30 seconds anyway, what harm does that cause, really
(especially if no-one asks about that name in the 30-second interval)?

> How would you like a car that drops a passanger whenever the loading
> on the engine goes above 90% ?

Depends on the passengers it's carrying :-)

Seriously, though, a DNS cache entry can be easily regenerated, so your analogy
is faulty.

> Caching is in the design for a reason. Systematically overriding it _is_
> breaking standards.

What standards?


- Kevin