In-Memory Knowledge Graphs – prepare, analyze and explore connected data at enterprise scale
Cambridge Semantics leads the market in connected data analytics on In-Memory Knowledge Graphs. Our breakthrough AnzoGraph™ is the most advanced of its kind – performing complex ad-hoc, OLAP interactive and batch queries at connected data scales and performance levels that our competition simply cannot match.
Unlike any other graph database technologies on the market, AnzoGraph was designed from the ground up to excel at “GOLAP” (Graph Online Analytics Processing) query style workloads. GOLAP is the name we have given to the class of connected data analytics that includes complex queries, often requiring many joins, filters and aggregation. These queries can be similar (although have the potential to be far more complex since we are dealing with far richer data models) to those OLAP (Online Analytical Processing) queries performed today in data warehouses but are now finally also available for connected data analytics too.
The difference between AnzoGraph and where the competition falls short is comparable to the difference between relational database technologies optimized for OLAP versus those tuned for OLTP (Online Transactional Processing). Other graph databases tend to do reasonably well at quickly locating a particular vertex in a graph and returning or updating whatever it is connected to, in other words a more transactional, record oriented pattern of access. Some of them even do reasonably well on queries that have a few joins, but none of them do well at rapidly loading a large amount of connected data and then performing complex combinations of join and filtering queries that touch far reaching corners of the graph or aggregation queries that require massive graph data traversals.
How it Works
AnzoGraph is simply and by far, the most performant Native Parallel Graph (NPG) analytics engine on the market, built to handle the wide variety and massive scale of today’s enterprise-wide data assets. It is an all in-memory Massively Parallel Processing (MPP) engine that delivers real time interactive performance for ad hoc queries, executing complex, many-hop “joins” and filters on the fly, to enable users to explore, pivot, and analyze data represented in entity rich models. AnzoGraph is capable of loading connected data at unprecedented ingest rates of between 2 to 4 million graph vertices/edges per second, per cluster compute node, resulting in very very large knowledge graphs being available to query in just minutes!
Once the data is loaded, AnzoGraph is able to query hundreds of billions of interconnected facts and relationships in real time without query tuning or indexing. More users, more data or more query throughput required? No problem, just add compute nodes. AnzoGraph is parallel and therefore scales linearly from a single server to clusters with hundreds of commodity server nodes. With this capability, Anzo 4.0 is proven to process analytic queries more than 100 times faster than its nearest competitor and at scale.
Graphmarts connect the power of AnzoGraph to the data in Anzo, bringing together subsets of data on demand for preparation, analysis and interactive access. Graphmarts are collections of data sets that can be shared, discovered and enhanced collaboratively. Anzo automatically brings Graphmarts quickly online into existing AnzoGraph clusters, or can provision new clusters on demand as business needs peak for agile deployment and cost effective cloud resource utilization.
While many users enjoy the visualization and analytic capabilities native to Anzo’s Hi-Res™ Analytics, Graphmarts support data-on-demand endpoints that allow users to off-ramp data from Anzo into their “last mile” analytics tools of choice through standard protocols like OData, REST and a novel DataFrame builder. Upon finding interesting multi-dimensional data sets through exploration of the Anzo catalog, users can use their favorite business intelligence or advanced analytics and machine learning tools to produce a final analysis and then share the results within their organization.
