RE: XML-*

Dave Pawson writes:

>> > From: noah_mendelsohn@us.ibm.com

>> > I'm not claiming that there are necessarily factors of, say, 
>> > 2x in all 
>> > this.  I'm saying that the harder you work on performance, 
>> > the more these 
>> > things start to matter. 

>>  So, at the extremes of efficiency, for one W3C 
>> area, we are considering removing something 
>> the remainder of the XML users may want?

>>  That sounds odd to me. Fix it for the 0.4% minority?

0.4% is your number.  SOAP is targeting those who want to use XML for 
messaging.  Of those, the number that want blazingly fast performance, or 
viewed another way those who are struggling to achieve the performance 
they used to have in binary now that they've switched to XML, is nothing 
like 0.4%.  I would guess (based soley on intuition) that it's 30-70% of 
the organizations with a serious interest in SOAP, and that they represent 
that vast majority of messages actually processed.

Let's say that the optimizations enabled by these optimizations represents 
20-30% of the performance that one would get in an otherwise highly tuned 
processor.  First of all, to a large customer, that might mean buying 8 or 
9 servers rather than 10 or 11 (noting that not all of the server load is 
in XML processing...though all too much of it is these days.)  That's a 
real tangible savings to users. 

At least as important is that aiming for this design point represents a 
philosophy to be applied consistently.   One almost never gets high 
performance with one optimization.  If my intuitive 20-30% number is right 
(and it may be a bit optimistic), that counts as a pretty big chunk in the 
performance optimization world.  Find another one or two such "wins" and 
the impacts really start to add up.    Two 30% wins and you've got a 
system that runs 2.5x faster. 

So, I think the tradeoff is an interesting one, but I don't think you've 
categorized it quite correctly. 

Some time ago, there was a proposal that any time an individual 
contributed 3 times or so on one issue on the tag list, that would be a 
reasonable limit.   I also think we've successfully aired for the tag's 
benefit a range of perspectives on the "should SOAP have done a subset" 
issue.   I plan to back off from this discussion unless something crops up 
where I feel I have a unique perspective.  Thanks.

------------------------------------------------------------------
Noah Mendelsohn                              Voice: 1-617-693-4036
IBM Corporation                                Fax: 1-617-693-8676
One Rogers Street
Cambridge, MA 02142
------------------------------------------------------------------

Received on Tuesday, 10 December 2002 12:10:26 UTC