[bind10-dev] b10-auth query performance: experiments and suggestions

[JINMEI Tatuya / 神明達哉]

At Thu, 19 Jan 2012 05:51:14 +0100,
Michal 'vorner' Vaner <michal.vaner at nic.cz> wrote:

> > We'll need something like these anyway, not for improving response
> > performance but for reducing memory footprint.  The current
> > implementation stored everything in the form of higher-level objects,
> > and very inefficient in terms of memory usage.  This change would also
> > be a necessary first step toward shared-memory type backend.
> 
> I fear that the shared memory backend would be more complicated than just this.
> As the memory will be at a difference place in the address space, we won't be
> able to use pointers. So it looks like we'll need to have some arrays with
> different kinds of data and index them or have offset pointers or something like
> that.

Right.  I didn't intend to claim the simple serialization was directly
applicable to the shared memory backend.  But I thought it was on the
way toward the right direction, and helped gradual and continuous
improvements rather than introducing one big change.

> Anyway, I'd like to propose not to do these optimisations of the tree before we
> design the new (shared memory) tree, unless we have a really good reason to do
> them now.

In any case I don't think we'll work on this second set of
optimizations soon.  It will be after year 3 at the earliest.  I'm
also okay with deferring it after the shared memory design if it can
be done soonish (e.g. as the first big feature of year 4), but, if
not, I think we should at least consider reducing memory footprint to
some reasonable level.  The one I showed in this thread is one way to
do that, but in that context I don't necessarily insist that's the
best one.

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