SingleStore

In addition to the tabular and vector formats already discussed, SingleStoreDB supports the JSON format, geospatial data types and operations, full text search, and time series. In the geospatial case, the product has data types for points, curves, and polygons, along with geographic indexing. Geographic features understand the curvature of the Earth but planar capabilities, for example mapping a factory floor, are not provided. On the time-series front, the product’s architecture is well-suited to the efficient storage of time-series data, and includes features like time buckets and interpolation.

Fig 2 - SingleStoreDB architecture

Moreover, SingleStoreDB accommodates both transactional and (real time) analytical workloads, effectively acting as either an operational database or a data warehouse depending on the task at hand. Analytics in particular are done with minimal, if any, data movement. What’s more, a (patented) single table type supports both transactional and analytical operations, allowing for universal data storage while enabling both types of workload. These operations are performed at speed using the product’s three-tier storage model, shown in Figure 2.

The product uses a distributed cost-based query optimiser that is aware of how data is stored to direct query operations to in-memory or disk-based storage as required. Note that SingleStoreDB uses skip-lists as its primary indexing technique, as well as supporting secondary indexes. Separation of storage and compute is another relevant scalability feature.

SingleStoreDB’s Workload Manager allows both memory and CPU resources to be allocated by user group, while its distributed ingest supports both batch and stream-based ingestion. Multi-cluster management is provided through SingleStore Studio, along with facilities such as encryption, security, role-based access control and auditing. There is also a SQL API that serves to support third-party environments such as Tableau and Qlik, and a specialised API – SingleStore Kai – designed to enable fast analytics for MongoDB applications. Finally, SingleStore has recently released SQrL, a chatbot tool that lets you sift through technical documentation or receive code samples via generative AI queries.

SingleStoreDB is a modern, distributed relational database and brings all the advantages you would expect from the familiarity of SQL. Wire compatibility with MySQL is a major advantage for anybody looking to migrate from an incumbent relational supplier, and the product reverses the use of memory and disk storage with respect to rows and columns, which means that you can get better performance for transaction processing while using fewer hardware resources and maintaining low latency for analytics and query processing. It also has strong geo-spatial and time-series capabilities.

What’s more, replacing a complex system with a simpler and more comprehensible one is often a good idea, and this is what SingleStoreDB allows you to do by subsuming the role of many more specialised offerings. Its support for vector data is particularly notable, as vectors have grown in popularity recently due to their perceived synergy with generative AI engines and the LLMs (Large Language Models) that power them.

Moreover, vectors are useful to LLMs because they can be used to supply important context. It stands to reason that this context is at its fullest, and therefore most effective, when created from a combination of vector and other data. A unified, multi-model database that includes vector data, such as SingleStoreDB, seems the easiest and most direct route to achieving this.

The Bottom Line

In our view, SingleStoreDB starts from the right place without being encumbered by historical design decisions, and the company has proven that it is a serious competitor to the more established relational database vendors. Its proposition of replacing many specialised databases with one, as well as its support for vectors, should both be highly appealing in the current data landscape.