John L. Myers has an interesting hypothesis for the future of NoSQL databases based
on their capability of handling “unstructured” data:
I think the future of NoSQL platforms is going to reside in the
ability of those systems to apply different operational or
analytical schemas to multi-structured data sets rather than letting
the data reside in a schema-free format. Merely storing multi-
structured data sets will not be enough to have a NoSQL platform
meet business objectives. The true business value will be in the
ability to apply the structures of a particular schema for analysis
or for operational workloads in real-time or near real-time.
What Myers suggests here is that storing unstructured data allows an application to define different “schemas” to repurpose the way data is used. In theory this sounds quite interesting. If done dynamically, this could define a system that could provide both OLTP and OLAP features.
The structure of the data has a very important influence on the data access implementation and the simple addition of structure metadata would not lead to the system to continue to perform optimally in various scenarios or for different workloads. Put it differently, OLTP and OLAP systems require data to be organized (and stored) differently in order to handle the different access patterns and different workloads. Switching from one to another while maintaining the characteristics of the system (reliability, performance, stability, etc.) seems to lead to a level of complexity that would be very difficult for a single system to handle.
Original title and link: The Future of NoSQL Databases: Hybrid Tools for OLTP and OLAP ( ©myNoSQL)