[bind10-dev] DNS packet API: the name object

[Jeremy C. Reed]

On Tue, 11 Aug 2009, JINMEI Tatuya / 神明達哉 wrote:

> > Speaking of benchmarking... do we have performance targets for this
> > class? If we intend to use it in packet processing, we can do some
> > guesses as to what is acceptable based on that.
> > 
> > So, for example, if we assume an authoritative-only server needs to
> > answer 100K queries a second on our as-yet-undefined target, and it
> > creates an average of 10 names, with 50 comparisons, then we would want
> > to be able to do something like 10M creations and 50M comparisons a
> > second on that machine if we can afford to spend 20% of our time doing
> > name processing. I have no idea if these numbers make sense, of
> > course. :)
> 
> Maybe we can begin with microbenchmark for basic operations such as
> from-wire, comparison, concatenation, split, etc, against BIND9 libdns
> and the BIND10 implementation.  Whether we need system-wide benchmark
> for this module (class) would depend on whether these operations are
> major bottleneck in the actual message processing.

...

> > We have Jeremy willing and able to help with documentation. Does it make
> > sense for him to start with documenting this class while you continue
> > working on the implementation? It would be great to have the two of you
> > co-operating on this.
> 
> I'll talk to Jeremy when he comes back from vacation.  I don't think
> we have to write a precise document of this prototype
> interface/implementation at this stage (it will surely change).  But
> we can use it as a practice of how to make a useful document while
> writing code.

Please let me know what you want from me. 

Also as I help write documentation, I can use the new interfaces to write 
some microbenchmarking tests.