Re:Building a scalable time-series database on PostgreSQL
by rodionos on Tuesday April 04, 2017 @04:47PM
(#14037041)
Reminds me of IBM Informix TimeSeries which has been around for a while. This was IBM's attempt to bridge the gap between the relational model and the columnar format that is optimal for time series, except that they store the time:value arrays in blobs. Aside from compression benefits the columnar storage works really well for irregular time series. If the incoming data is irregularly spaced, row-based schema will have a certain degree of inefficiency due to the need to store NULLs each time one of the columns is missing a valid value.
[ Parent
| on HN ]
Re:Building a scalable time-series database on PostgreSQL
by enordstr on Wednesday April 05, 2017 @08:33AM
(#14041591)
That point about irregularly spaced data (sparse) is a very insightful observation. I’d just add that a user can to some extent address that by normalization, i.e., splitting incoming data across multiple TimescaleDB (hyper)tables, like in any SQL database. However, the are clear trade-offs here. The upside is that users can themselves balance these trade-offs.