关于sqlserver xml使用上的一些建议(摘录于sqlserver bible 2k8)

XML should be used only where it is really required. Using XML where relational data would
best be suited is not a good idea. Similarly, using a relational model where XML might run
better won’t produce the desired results.

XML is good for storing semi-structured or unstructured data. XML is a better choice if the
physical order of values is significant and the data represents a hierarchy. If the values are
valid XML documents and need to be queried, storing them on an XML column will be a
better choice over VARCHAR, NVARCHAR, or VARBINARY columns.

If the structure of the XML documents is defined, using typed XML columns will be a better
choice. Typed XML columns provide better metadata information and allow SQL Server to
optimize queries running over typed XML columns. Furthermore, typed XML provides storage
optimization and static type checking.

Creating a primary XML index and a secondary XML index (or more, depending on the workload)
might help improve XQuery performance. An XML primary index usually uses up to
three times the storage space than the data in the base table. This indicates that, just like SQL
indexes, XML indexes also should be used wisely. Keep in mind that a full-text index can be
created on an XML column. A wise combination of a full-text index with XML indexes might
be a better choice in many situations.

Creating property tables to promote multi-valued properties from the XML column may be
a good idea in many cases. One or more property tables may be created from the data in an
XML column, and these tables can be indexed to improve performance further.

Two common mistakes that add a lot of overhead to XQuery processing are usage of wildcards
in the path expression and using a parent node accessor to read information from upper-level
nodes.

Using specific markups instead of generic markups will enhance performance significantly.
Generic markups do not perform well and do not allow XML index lookups to be done
efficiently.

Attribute-centric markup is a better choice than element-centric markup. Processing information
from attributes is much more efficient than processing information from elements.
Attribute-centric markups take less storage space than element-centric markups, and the
evaluation of predicates is more efficient because the attribute’s value is stored in the same row
as its markup in the primary XML index.

An in-place update of the XML data type gives better performance in most cases. If the update
operation requires modifying the value of one or more elements or attributes, it is a better
practice to modify those elements and attributes using XML DML functions, rather than
replace the whole document.

Using the exist() method to check for the existence of a value is much more efficient than
using the value() method. Parameterizing XQuery and XML DML expressions is much more
efficient than executing dynamic SQL statements.