I have finished (for now anyway) my Xalan performance analysis and have determined that the DTM code is responsible for making Xalan unusable in our production environment. I tracked down several data structures in the DTM classes, some of which I reported in an earlier email, that are initialized to large sizes and are duplicated many times during a transformation. The memory consumption is so excessive that when running concurrent transformations it completely overwhelms the machine.
Has there been any analysis that shows a big improvement with the DTM model, or does it still need to be tweeked to get it working efficiently? I hope someone has done, or plans to do, some evaluation of the code to make sure things are improving for someone. Its unfortunate that version 2.1.0 is the end of the line for us after using Xalan very successfully in production for over a year now. Now I have to find a suitable replacement. Chris
