Generally speaking, there are two programming models for working with XML infosets: streaming and the document object model (DOM).
The DOM model involves creating in-memory objects representing an entire document tree and the complete infoset state for an XML document. Once in memory, DOM trees can be navigated freely and parsed arbitrarily, and as such provide maximum flexibility for developers. However, the cost of this flexibility is a potentially large memory footprint and significant processor requirements, because the entire representation of the document must be held in memory as objects for the duration of the document processing. This may not be an issue when working with small documents, but memory and processor requirements can escalate quickly with document size.
Streaming refers to a programming model in which XML infosets are transmitted and parsed serially at application runtime, often in real time, and often from dynamic sources whose contents are not precisely known beforehand. Moreover, stream-based parsers can start generating output immediately, and infoset elements can be discarded and garbage collected immediately after they are used. While providing a smaller memory footprint, reduced processor requirements, and higher performance in certain situations, the primary trade-off with stream processing is that you can only see the infoset state at one location at a time in the document. You are essentially limited to the “cardboard tube” view of a document, the implication being that you need to know what processing you want to do before reading the XML document.
Streaming models for XML processing are particularly useful when your application has strict memory limitations, as with a cell phone running J2ME, or when your application needs to simultaneously process several requests, as with an application server. In fact, it can be argued that the majority of XML business logic can benefit from stream processing, and does not require the in-memory maintenance of entire DOM trees.