Have you guys had any interest from the existing SCADA process historian vendors (Wonderware Historian, OSISoft PI, etc) or historian users about using Timescale?
We use process historians, and it's unlikely we will ever get off them. Industrial process historians are hard to replace because of the multitude of corner conditions that they handle, a result of being hardened over 3-4 decades of being deployed in industry.
That said, most process historians are really not optimized for analytics use cases, which are becoming increasingly important for extracting value. They also don't scale too well horizontally, and are unwieldy to work with on the large. (typically they're used for visualization, and you are typically only able to work with small subsets of the data at a time, like when you're troubleshooting operations at a particular plant, as opposed to finding patterns across all plants which is more easily achieved with something like Timescale).
I think an ideal setup for me would be to get historian collectors to tee their data to both the historian database, and to something like Timescale. I do a lot of large scale time-series modeling and would really prefer to work with a SQL database than a historian API.
wenc's comments are spot on. Just to add: we have found ourselves being used alongside historians, as an easier way to extract and analyze the actual data.
In other words, instead of "rip and replace" companies are using TimescaleDB as a more flexible, powerful and easier-to-use operational analytical engine alongside historians, for easier access to their data via full-SQL and in a relational model that lends itself to more useful, complex analysis.
Always intrigued to learn more about where we can help here, so if you have any insight, we'd love to hear from you: hello@timescale.com
I think the scada vendors have already solved the same set of problems that timescale solves.
Edit: pretty sure the scada vendors partition the data in to tables so each table is a day or week of data, they do rollups on the data so the detail is available if you do a query with a 1 minute timespan but the data is already aggregated in to representative values for a 1 minute interval if the query timespan is 1 day.
Sounds like timescales clustering story might be better, but with OSI or wonderware your clustering story is ms SQL or oracle.
Except that most process historians store data in a non-relational way which makes it difficult to integrate with other types of data.
Process historians, like most NoSQL time-series offerings, are traditionally optimized for fast data retrieval (for HMIs, real-time monitoring dashboards, etc.) but are really weak when it comes to analytics. There are new products like Seeq (a kind of a Tableau for process historians) that attempt to address this, but somehow miss the fact that for analytics, you really need your data to exist in a dataframe-like shape.
Today, if you want to do any kind of non-trivial analytics, you typically have to first ETL data from the historian to a relational database via REST or OLE DB API. TimescaleDB on the other hand already stores data in a native relational format and lets you run full SQL on it.
> pretty sure the scada vendors partition the data in to tables so each table is a day or week of data, they do rollups on the data so the detail is available if you do a query with a 1 minute timespan but the data is already aggregated in to representative values for a 1 minute interval if the query timespan is 1 day.
I have not seen this, but if this is the case, and if these aggregates are stored in a proper relational database, that solves a big chunk of the problems people have with doing analytics on historian data.
As far as I know GE Proficy does not do this. Which is a pain.
