From: Wolfgang Meier <sbhati@we...> - 2001-11-10 22:30:12
I just released a preliminary version of eXist 0.7 on sourceforge:
I'm quite happy about this release because it fixes a number of problems I
struggled with during the last two weeks.
The native Java backend is now safe for concurrent write access by
multiple client threads.
The performance breakdown which occured with previous versions of the
native storage backend when indexing many small files has been
fixed. The indexing process scales much better now with collections
many small files and queries perform well.
To achieve this, the indexing structure had to be changed. Please note
that the element-index is now organized on a collection basis. If
there are too many files in one collection, it is recommended to split
them up into subcollections.
Attribute values are now included in the fulltext index. An attribute
value may be queried with &= and |= operators.
The number of bytes used to store a node in the main dom file
has been heavily reduced.
The index used for string comparisons (=,<,>) has been
removed. Building this index is expensive. You may use the &= and |=
operators instead if a term based search is sufficient for you.
=,<,>,contains() etc. do a full table scan. Because of the reduced dom
storage size, full
table scans are much faster now.
Important: The configuration file contains some settings for the different
page buffer sizes in the db-connection
entry. If buffer sizes are set too high you will experience a considerable
I'm still trying to find a good average setting.
Norman Walsh's resolver classes are used in the parser to read XML
Catalog files and resolve entities.
Default namespaces (xmlns=) do no longer produce
xml-prefixes are handled correctly (e.g. xml:lang="...").