jexp / affinity-graph-comparison

Affinity Graph Comparison Code

This project contains the code we used to benchmark the various datastores we were considering for an affinity graph system. The actual benchmarking project is graphstore-load-test. The other five graphstore projects are the drivers used to interface with the datastores. The last project is a bulk loader for Elasticsearch, which we used to more efficiently load large datasets.

In our affinity graph system you have three entities; products, users and affinities, where affinities relate a user to a product. For example; user A likes product B. User A and Product B are the entities, and 'like' is the affinity.

While this sounds like an easy relational problem, the twist is that we want to support arbitrary key/value pairs associated with both products and users, and query on those key/value pairs. For example the system should be able to satisfy a query looking for any red haired users that like blu-ray products. Where red hair is an attribute of user, and blu-ray is an attribute on product.

Each sub-project in this folder contains a different 'driver' for each particular storage technology. In addition there is the graphstore-load-test directory which contains our testing system that we used to run queries against each driver to gauge their efficiency.