> -----Original Message-----
> From: Joseph Kesselman [mailto:[EMAIL PROTECTED] 
> Sent: Monday, September 27, 2004 5:32 PM
> To: [EMAIL PROTECTED]
> Cc: [EMAIL PROTECTED]
> Subject: Re: Piped transformations and multithreading
> 
> 
> 
> 
> 
> At this time, we do not optimize through pipes. Each stage produces an
> output document which is then taken as an input document by the next
> transformation. We may be able to eliminate some latency by running in
> "incremental" mode, but we don't eliminate any work except 
> that of going to
> and from XML syntax.
> 
> Avoiding complete construction of the input document model, for either
> pipelining *or* single transformation, is a Known Opportunity For
> Optimization -- but a nontrivial task. See archives for handwaving
> speculation about "filtering" and "pruning" based on knowledge of the
> stylesheet's requirements (and, ideally, of the document's schema). We
> believe it's possible; we do not currently have an 
> implementation which
> does it.
> 
> Even in incremental mode, Xalan is a pull-processor, not a 
> push-processor;
> if you hand us a push source such as a SAX stream we "throttle" it to
> accept tokens at the rate we're comfortable with -- and yes, 
> that further
> limits how much we can parallelize stages in a pipe. You could
> theoretically buffer the SAX stream in a FIFO queue to avoid 
> that lockstep,
> but that of course has its own costs.
> 
> 
> 
> ______________________________________
> Joe Kesselman, IBM Next-Generation Web Technologies: XML, XSL 
> and more.
> "The world changed profoundly and unpredictably the day Tim 
> Berners Lee
> got bitten by a radioactive spider." -- Rafe Culpin, in r.m.filk
> 
> 


So is the intermediate document (between pipeline stages) at least
optimal 
for Xalan?  i.e. a DTM? 
--------------------------------------------------------
 
NOTICE: If received in error, please destroy and notify sender.  Sender does not waive 
confidentiality or privilege, and use is prohibited. 
 

---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to