[XML-DEV Mailing List Archive Home] [By Thread] [By Date] [Recent Entries] [Reply To This Message]

RE: Parsing Performance

• To: "'David Megginson'" <david@m...>,<xml-dev@l...>
• Subject: RE: Parsing Performance
• From: "Martin Soukup" <martin@d...>
• Date: Sat, 22 Feb 2003 23:23:29 -0500
• Importance: Normal
• In-reply-to: <15959.37012.762610.653552@m...>

>  > I did the same in Java, and it is well over 1000 times slower. I
>  > agree with John that performance isn't everything, but 1000 times
>  > slower using the same technology is, IMO, completely unreasonable.
> 
> The biggest mistake people make designing XML processing systems is
> not distinguishing between what happens in the loop and what happens
> outside the loop.
> 
> Optimizations inside the parsing/processing loop can give enormous
> payoffs -- for example, if you have a method that's invoked for every
> element, every text chunk, or even every character, it's an excellent
> candidate for optimization.  On the other hand, optimizing a method
> that's invoked only once per document (unless your system handles
> thousands of small documents per minute) or once per session is
> generally a waste.
> 
> For example, let's say that a method currently runs in 10
> microseconds, and through brilliant refactoring, you make it 1000
> times faster.  If that method is invoked once for every character in a
> 1GB document, then you've just shaved 10 seconds off execution time --
> if you managed that with only a day's work, you deserve a productivity
> aware.  If, on the other hand, that method is invoked once during the
> session, you've shaved about 0.00001 seconds off execution time -- if
> you spent a whole day doing that, you deserve something more along the
> lines of a bozo award.
> 
> Using an exception to terminate parsing falls into the latter category
> unless, as I mentioned, you have an application that processes
> thousands of XML documents every minute.

Absolutely agree. Just that my application happens to do that :)

The biggest problem with large open standards is that they must serve a
wide community of users and through that an even wider range of
customers. I guess the issue here is that one should have the option of
doing things either way. Wouldn't that serve the most people?

• Follow-Ups:
  • RE: Parsing Performance
    • From: David Megginson <david@m...>
  • Re: Parsing Performance
    • From: Bill de hÓra <bill@d...>

• References:
  • Parsing Performance
    • From: David Megginson <david@m...>

• Prev by Date: Re: The subsetting has begun
• Next by Date: Re: The waterfall model lives? (was Re: The subsetting has begun)
• Previous by thread: Parsing Performance
• Next by thread: Re: Parsing Performance
• Index(es):
  • Date
  • Thread