I am trying to understand whether neo4j will work for my application. I am using it to record events overtime for example who votes in a particular way during a Brexit vote which I understand is a good use case for Neo but I have time series quantitative data like monthly inflation or GDP or quarterly industrial output for the entire period in question, which I want to then firstly import and visualize in relation to particular votes but also to run an algorithm that may examine if GDP growth was lower then expected by a particular threshold did this shift the votes of a particular group. Is this second example a good use case for neo and what would the data structure of monthly quantitative data eg GDP look like? Or could someone provide me a suggestion on how else to handle this problem?
massive time series data is not what Neo4j is built for but if it's just a bunch of measurements then you can just create them with date/datetime properties and index those properties and then link them to other information you have
For thresholds and such you might need to use aggregations (on differnt timeframes) and post filtering depending on what exactly you want to query.
Probably best to share more details / your model.
You might want to consider GraphAware Time Series Library
Hi @sommer , for timeseries data, the best choice it go for columnar databases like Cassandra (or datastax). Modelling with graphware maybe good for representation, but not for storing and fast retrieval. Also graphware is not supported in neo4j 4.x.
For example, if you have 10 sensors and each sensors have metrics with datetime, then you query those 1 of the sensors , then by datetime and then their metrics.
Cassandra has both availability and partition, whereas Neo4j has availability and consistency.
This is the database we are using currently -> https://www.influxdata.com/
Actually graphaware updated their framework for Neo4j 4.x
I think it all depends on the volume, if it's just the GDPs of all countries on a monthly basis, neo4j is totally fine.
If it's billions of sensor events per hour or day then perhaps not.