|
Hi,
We are using Xerces C++ very successfully in a
real-time transaction switching application. Performance is excellent, but one
drawback seems to be memory fragmentation behaviour resulting in continuous
arena growth that seems to be caused by the document-based memory management
implementation.
I was wondering if you would consider incorporation
of changes to move memory management up to the parser so that the parser dishes
out memory blocks for document parses - as an option during compilation. For
constructed documents you would then also have to supply a parser to be used to
dish out memory.
The parser will then keep a high water mark and the
allocated memory area gets reset in the ResetDocumentPool method.
We have implemented these changes in Xerces 1.6 and
will now redo on Xerces 2.x. These changes worked well, but some better design
may prove beneficial as it does not fit to pass a parser object to a document
being construcuted. Perhaps a separate memory manager class is required with
methods that allows setting of allocation chunk sizes etc. - this can then be
supplied to documents being constructed and to Prser objects for use during
document parsing.
I may also be missing something
altogether?
An algorithm like this will relieve significant
strain on applications parsing documents at a very high rate and that typically
processes a lot of the same type & size documents as is the case in our
transactional system.
Your feedback would be much
appreciated.
Regards
Philip Stander
|
