home ¦ Archives ¦ Atom ¦ RSS

Big Dicts

atbr seems like an interesting approach to really large scale in memory key/value store, otherwise known as dictionaries, or dicts, in Python.

…atbr is basically a thin swig-wrapper around Google’s (memory efficient) opensource sparsehash (written in C++). Atbr also supports relatively efficient loading of tsv key value files (tab separated files) since loading mapreduce output data quickly is one of our main use cases.

While the authors seem a little more focused on Hadoop integration, I’ve got another interesting use case. NetworkX is a well developed Python module for graph representation, manipulation, and algorithms. The module uses Python’s built-in dicts as the primary data structure to represent these graphs. In my experience, NetworkX tends to fall over a bit with big graphs. Maybe using atbr as a replacement underneath NetworkX would improve both memory usage and execution speed. Yet another personal hacking project I could adopt.

Also of interest was some of the benchmarking that inspired atbr and demonstrated that Python dicts are actually pretty decent.

© 2008 ‐ 2023 C. Ross Jam. Built using Pelican. Theme based upon Giulio Fidente’s original svbhack, and slightly modified by crossjam.