Currently, the dmrpp module treats single chunk data items as if they are the same as a multi-chunk item. This means all of the machinery for parallel access and assembly are employed in what is really a simple read operation. We have code to read 'contiguous' data, but it only works for those things that HDF5 calls contiguous - meaning not compressed. Single chunk data items are just like HDF5 contiguous things, except they can be compressed (or otherwise filtered). We can improve performance for these single-chunk items by adding support for compression/filters to the read_contiguous() code.