Release of the GCX XQuery EngineQ

[Jonathan Robie]

Stefanie Scherzinger wrote:
>> In order to be a little more concrete, I talked with Marc about some
>> comparisons we did internally between GCX and DataDirect XQuery. I think
>> the results fall out roughly as follows:
>>
>> - Our XMark performance results are broadly similar to those of GCX for
>> our currently released software. The upcoming DataDirect XQuery 3.0 will
>> be markedly faster in some cases - for instance, for one XMark query, we
>> finished more than 100 times faster.
>
> I assume that the one XMark query where you outperform us in runtime
> is the one with the join, where we still only have a naive nested-loop
> implementation.

Yes, Q8.

>> - Our memory usage was similar for many queries, worse for some, and
>> better for others.
>
> This is very interesting, do you think we could include experiments
> with your product on our benchmark site?

Not my call - I'm afraid we're just big enough that our marketing folks 
would need to be in on that decision. I'll copy this to them and 
encourage them to say yes - let's take this part  to private conversaton.

>> As Marc indicated, we support both document projection and streaming (as
>> described in A. Marian and J. Simeon, "Projecting XML Documents", Proc.
>> VLDB 2003, a really cool and useful paper).
>
> Out of curiosity - does your projection strictly adhere to the
> technique described in this paper? GCX projection is similar, but it
> is also capable of omitting inner nodes. For instance, to project for
> the path //a in <x><b><a/></b></x>, the x- and b-labeled nodes are not
> buffered by GCX. Interestingly, this can really make a difference when
> projecting very large XML documents, even if they are shallow.

I am not aware that we eliminate inner nodes, but Marc knows the details 
better than I do. I'm surprised that it would make such a difference, 
since there are generally many fewer outer nodes than inner nodes.

Jonathan