Trovares xGT

Trovares xGT is a property graph engine. It is not a graph database per se. However, it will compete with graph databases for query processing against extremely large graphs. Briefly, the concept is that you load your data – currently it must be in CSV format – into memory. This is done in parallel so that ingestion rates are rapid. The software will convert the loaded data into graph format, ready for exploration. There is a free trial version of the product available on AWS (Azure in progress) with anything from 1Gb to 4Tb on an hourly basis. Or you can arrange an on-site proof of concept.

Customer Quote

“Trovares is uniquely scalable on extremely large datasets. We leverage Trovares to identify advanced malware behaviors in massive network systems with record ingest rates and scalability.”
Neo Prime Inc

Trovares xGT is all about performance and scale. It has demonstrated the ability to support a 1 trillion edge graph though this was on a clustered solution that is not yet commercially available. In its initial incarnation xGT is marketed in a single node configuration; capable of supporting tens of billions of edges in a 4Tb instance on AWS.

As far as processing is concerned, xGT supports both Python scripts and TQL (Trovares Query Language), which is a subset of OpenCypher. As an example of what is not supported by TQL, Cypher allows you to add edges within a query, but this is not supported by TQL. Over time the company expects TQL to get closer to the OpenCypher standard. The company is working on the provision of graph algorithms, such as PageRank, but these will not be available until 2019. Also in 2019, the company expects to be introducing a cost-based optimiser.

Given the early nature of xGT it is not surprising that the company hasfew partners with complementary technologies. However, Trovares has teamed with Neo Prime, a Silicon Valley cyber and quantitative risk modeling firm, to power Neo Prime’s risk-driven behavioral query engine to detect malware behaviors in massive network systems. At present, there is no formal way of integrating with visualisation, data quality or data integration tools: users are expected to write Python scripts to support such integrations.

Trovares describes the graph problem statement as “gigabyte scale tools for petabyte size problems”. It argues, with some justification, that if you really want to get performance and scale for graph analysis, you need to build parallelism into the product from the outset, rather than try to retrofit that later. We agree with that proposition and xGT certainly has potential. However, it is currently restricted to single node implementations and has a limited query optimiser. This gives the product lots of scope to improve over time but limits its claims to scale and performance for the present. Moreover, although the company has done some benchmark testing in which it performs well, the competitor it chose to benchmark against is unlikely to be its main rival for super-large, complex analytics.

The Bottom Line

Trovares xGT is interesting and it is certainly worth investigating for suitable use cases. As a very immature product we are not in a position to recommend it at this stage, though we will watch the company’s progress with interest.