[bind10-dev] Profiling & performance sugestions

[Michal 'vorner' Vaner]

Hello

On Mon, Jan 24, 2011 at 11:12:43AM -0800, JINMEI Tatuya / 神明達哉 wrote:
> At Mon, 24 Jan 2011 13:30:07 +0100,
> Shane Kerr <shane at isc.org> wrote:
> 
> > I hope we have some time on Wednesday's call to discuss this issue. If
> > not, perhaps it would make sense to set up a separate meeting for this
> > topic (either in voice or chat, I'm fine with anything)?
> 
> I love performance improvement work, and discussing ideas of
> optimization is a good thing, but I'd rather avoid *doing* premature
> optimization.  The in-memory based authoritative server is still
> feature incomplete, and completing the feature may reveal more severe
> bottlenecks or may make prematurely introduced optimizations marginal.

Well, if we need the performance in some 6 weeks and we expect to get CNAME,
DNAME and wildcard, I don't think it is premature. If you look at it, there are
some major bottlenecks (and I have no idea how we could make these points
marginal, in a different way than introducing something a lot slower). Some of
it is name compression, which we will need as well. Other is traversing the
tree, which will be there. The creation and destruction of many UDPServer
objects will be valid until we fix it or refactore the processing, which is not
planned.

The fact that OutputBuffer::operator[] is a hot code and Name class is a hot
code is hard to believe to change. If we can optimise them inside without
changing the interface and without much effort, then I don't see much problem in
it.

I agree that, for example, the shared pointers, memory pools and stuff might be
too soon (or, it might be too soon until the end of Y2 at last).

But if we agree that what we currently have is slow and needs to be sped up,
then this is the time we at last start talking and experimenting, at last a
little bit.

Have a nice day

-- 
do { goto Water; } while( !tryBreakOffTheHandle() );

Michal 'vorner' Vaner
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <https://lists.isc.org/pipermail/bind10-dev/attachments/20110124/2d1c021c/attachment.bin>