Hi @rkumar,
There should be no obvious steps you need to explicitly take to manage data resources. Given we have no knowledge of your complete application nor visibility within the processing of data, the best way to eliminate concerns specifically related to opening multiple streams is to do this within a standalone application. Try opening multiple streams within one of the training examples and observe the behaviour. If you observe the out-of-memory behaviour with the example, then you can report your findings here.
