[xsde-users] Question on the XSD/e performance

Boris Kolpackov boris at codesynthesis.com
Thu Jan 3 08:00:32 EST 2008

Hi Mahesh,

Mahesh Kelkar <mkelkar at rocketmail.com> writes:

> Is performance of the XSD/e limited by the underlying parser performance?

Yes, it is currently limited by the underlying (non-validating) XML parser

> Or as a part of C++ code generation do you modify any parser code to
> streamline the code execution?

The data extraction, XML Schema validation, and dispatching code
is generated based on a particular schema which results in better
performance compared to the general-purpose validating XML parsers.
There is not really much to be gained from generating custom, low-
level XML parser (also see notes below).

> Do you have any plans to improve the performance of the overall
> package by adopting recommendations of the XML-scremer?

We do not have any immediate plans to do so. As mentioned in
the paper written based on this research project, many of the
optimization techniques employed by XML Screamer are only
applicable when certain assumptions are made about XML documents
being processed. In other words these techniques only work on a
subset of XML and are not possible (or hard to implement) in the
general case.

Having said that, if your XML documents are not using certain
expensive (from the parsing performance point of view) features
(e.g., entity references), then it may be possible to come up
with a custom underlying XML parser for this XML subset with
a much better performance compared to general-purpose XML
parsers that must be able to parse all conforming XML documents.

Another approach that proved to work very well for some of our
clients is to use binary instead of textual representation of
XML. For example, you could use a binary format for internal
interchange/storage in your application and normal text XML for
interoperability with outside world. XSD can produce code from
XML Schema for parsing/serializing in several binary formats
with typical parsing speeds being about 10 times faster than
text XML parsing.


More information about the xsde-users mailing list