Query: Libxml performance on multiprocessor system

Hi,

   We tried to get the performance of Libxml (version-2-2.6.23) on a multiprocessor environment, with two threads running simultaneosly [ Kernel level threads ].

   And were expecting that the performance will scale up, but instead the performance recorded is almost similar [3% more] to what we get for single thread.

 

   Each thread parses the xml document using xmlSAXUserParserMemory.

 

   Can anyone help me to identify the root cause for this??

   I surmise that this may be because of the extensive locking of Data and/or Api's done inside Libxml.

 

   Do we have any data for multithreaded performance on a multiprocessor system??

  

  Also, can anyone tell that to what extent the performance of Libxml is expected to scale up on a multiprocessor environment.

Thanks in advance !! 

Regards,

Harbhanu