RE: Performance of DOM/SAXSource vs. StreamSource & other optimization questions

I was referring to caching the source of the stylesheet, not the input.

In any case, I'll start down the road of caching the Templates object,
then look at translets if I need more oomph.

Thanks.

> -----Original Message-----
> From: Hall, Joe [mailto:Joe.Hall@xxxxxxxxx] 
> Sent: Friday, September 09, 2005 1:41 PM
> To: Karr, David; xalan-j-users@xxxxxxxxxxxxxx
> Subject: RE: Performance of DOM/SAXSource vs. StreamSource & 
> other optimization questions
> 
> 
> We cache the Templates object.  We then took this a step further and
> started using translets.   We noticed a marked performance improvement
> by caching the Templates and using translets.
> 
> http://xml.apache.org/xalan-j/xsltc_usage.html
> 
> It was impossible for us to cache the source since it changed 
> every transformation.  Does your input remain the same over 
> several invocations?  If it does then you should cache the 
> result of the transformation and use that when you don't have 
> new input content.
> 
> -Joe
> 
> -----Original Message-----
> From: Karr, David [mailto:david.karr@xxxxxxxx] 
> Sent: Friday, September 09, 2005 3:57 PM
> To: xalan-j-users@xxxxxxxxxxxxxx
> Subject: Performance of DOM/SAXSource vs. StreamSource & 
> other optimization questions
> 
> I haven't tried any of this yet, but I'm investigating 
> options for optimizing XSLT.  Caching the Transformer object 
> is attractive, but synchronization issues will make this less 
> attractive.  I'm looking at implementing this in a messaging 
> router, which could concievably be attempting the same 
> conversion on 10 different threads at about the same time, 
> and doing that many thousands of times in a day.
> 
> I'm wondering instead about caching a Source object.  Is it 
> practical to do this?  What about caching a DOMSource or 
> SAXSource instead?  I would guess that would be better than a 
> plain StreamSource object.  None of these classes say they 
> are NOT thread-safe, so I assume they are ok with concurrent 
> references (although I find that a little hard to believe).
> 
> This would be implemented with JDK 1.4.2.  Would the "Source" 
> features be enabled in the stock JAXP implementation in that version?
> 
> What does the "optimize" attribute in Xalan do?  Is that 
> available in the stock JAXP implementation?
>