[bind10-dev] Question about statistics incremental update

[JINMEI Tatuya / 神明達哉]

At Tue, 9 Oct 2012 15:02:46 +0200,
Michal 'vorner' Vaner <michal.vaner at nic.cz> wrote:

> > Question 1: Which do you prefer that transfered statistics data
> >             is increment(delta) or counter data itself ?
> > 
> > * Option 1: some of statistics data are increment(delta).
> >             each statistics data is an increment from last read.
> >             it is transfered if the increment from last read is not zero.
> > 
> > * Option 2: Each statistics data is transfered
> >             if the data is changed frm last read.
> > 
> >             pros: It is already implemented at trac2179:
> >                   Update Stats to support partial statistics updates
> > 
> >             cons: It requires previous counters at last read
> >                   and it doubles couter memory usage at statistics targets.
> 
> Personally, I like the first option more. For one, the memory consumption
> (actually, we may be unaware of the list of zones before they are queried, if
> there are really many zones in the database, the option 1 allows us to forget
> about them again when sent ‒ with option 2, the list of seldom used zones must
> still be kept in memory).

I prefer option 1, too.  I'd like to keep the max amount of
state/memory footprint for auth as predictable as possible, and it's
better if the auth server doesn't have to maintain mostly unnecessary
state.

As for other questions, I'm not sure.  Maybe I really don't understand
the points, but these seem to be relatively minor details and I'm
afraid I cannot say anything useful unless I see near-complete
implementation.

---
JINMEI, Tatuya
Internet Systems Consortium, Inc.