Rockset vs. SingleStore
What is Rockset?
According to its website, Rockset is a “search and analytics database built for the cloud, with real-time indexing and full-featured SQL on JSON, time series, geospatial and vector data”. Rockset is built on RocksDB, and is primarily a document store with SQL being a secondary data model.
What is SingleStore?
SingleStore is the world's only real-time data platform. It can read, write and reason on petabyte-scale data in a few milliseconds. At its core is a high-performance, SQL database engine that allows you to run transactions, analytics and vector search— eliminating the need for multiple databases and unnecessary data movement. SingleStore delivers high throughput ingestion from distributed data sources while serving single-digit millisecond queries, making it an ideal platform for real-time analytics and gen AI.
Fast data ingestion
Real-time analytics using SQL queries
Multimodel database with support for JSON, time series, geospatial and vector data
Stability at scale. Some users report that Rockset is unstable and buggy [source: publicly available employee reviews], as well as OOM issues for OLAP queries with large joins on large data sets [source: Reddit thread: Anyone using Rockset as their primary warehouse?]
No transaction support. Rockset is not a transactional database and is not designed to support OLTP workloads. Its architecture only supports eventual consistency; transactions lack Atomicity (the A in ACID) that are crucial for mission-critical applications.
Limited vector search capabilities. Indexed vector search in Rockset requires building a vector index for each partition, making it difficult to scale when dealing with changing datasets. The approach is also inefficient because it involves gathering a set of centroids close to the target vector, then applying a predicate to check if each row belongs to those nearby centroids. Indexed vector search is restricted to IVF algorithms and does not extend to more advanced algorithms like HNSW, which offer superior performance and recall.
Deployment. Only available on five cloud regions- All on AWS. There is no self-managed offering for users who may want to control performance, storage or security more extensively, or run on existing infrastructure.
Ease of getting started. No native support for Notebooks or co-pilot for writing code. No locally hosted offering.
Ingestion performance. Parallel high-throughput parallel ingestion — up to millions of events per second
Analytical performance. High performance and scalability with millisecond query response times; supports complex queries
Vector search. Support for both KNN and ANN and makes it easy for developers to do vector and full-text search with SQL.
CRUD. SingleStore is suitable for both OLTP and OLAP workloads. It stores data in patented row and column-based storage, making it extremely capable for both transactional and analytics use cases. SingleStore’s columnstore supports point updates, row-level locking and high-volume concurrent writes, upserts and deletes.
Multi-model. Support for SQL, JSON/BSON (MongoDB® API compatible), geospatial, key-value and time-series.
Bottomless storage and distributed architecture. Three-tier storage architecture with in-memory rowstore, on-disk columnstore and cloud object storage enabling separation of compute and storage; shared-nothing architecture that allows scaling to thousands of concurrent user sessions.
Deployment. Fully managed platform-as-a-service on all three hyperscaler clouds (AWS, Azure, GCP), and self-managed offering for Kubernetes/on-prem deployment
Data platform. Includes native data integration, compute services and Notebooks (SQL/Python/R).
Not ideal for pure transactional workloads (with requirements like foreign-keys, multi-master or geo-partitioning)
Not ideal for purely batch analytics
Not ideal for very small scale datasets (<1 GB)
Technical comparison
Speed
Ingestion and query performance
| Capability | Rockset | SingleStore |
|---|---|---|
| Ingestion performance | ⬤⬤⬤⬤ | ⬤⬤⬤⬤ |
| Rockset is capable of fast ingestion | SingleStore Pipelines offer streaming, parallelized data ingestion with optional transforms from multiple data sources such as Amazon S3, Kafka, HDFS and Iceberg. Freshly ingested data is immediately queryable. | |
| Query performance | ⬤⬤⬤ | ⬤⬤⬤⬤ |
| Rockset claims to support p95 query latency of about 70 ms | SingleStore supports low-latency analytics (~10s of milliseconds) on complex queries (involving aggregates, joins, filters, etc.) | |
| Indexes | ⬤⬤⬤ | ⬤⬤⬤⬤ |
| Rockset relies on a default inverted index (Converged Index) | SingleStore supports skiplist index, columnstore index, hash index and multiple indexes for vector search (Inverted File Index, HNSW with quantization techniques) |
Scale
Scale applications cost-efficiently
| Capability | Rockset | SingleStore |
|---|---|---|
| Separation of storage and compute | ⬤⬤⬤ | ⬤⬤⬤ |
| Rockset supports separation of storage and compute | SingleStore's three-tiered "bottomless storage" includes an in-memory rowstore, a disk-based columnstore (persistent cache) and cloud object storage. This separation of storage and compute enables supporting growing data volumes using low-cost storage infrastructure. | |
| Read Replicas | ⬤⬤⬤⬤ | ⬤⬤⬤⬤ |
| Rockset supports "compute-compute separation" to allow independent scaling of compute for multiple applications and workloads | SingleStore compute workspaces allow independently scaling compute for different workloads while utilizing a shared database. | |
| Concurrency | ⬤⬤⬤⬤ | |
| Not enough is publicly known about concurrency supported by Rockset | SingleStore's distributed SQL architecture can scale to thousands of concurrent user sessions | |
| Auto scaling | ⬤⬤ | ⬤⬤⬤ |
| Rockset's Virtual Instance Auto-Scaling is in Beta | SingleStore's auto-scaling allows you to quickly and automatically adjust compute resources based on changing workload demands. | |
| Compute service for ML | ⭘ | ⬤⬤ |
| Rockset does not include a native compute service | SingleStore's native compute service [In private preview] allows you to deploy scalable compute in a secure manner right next to your data-intensive application for AI/ML/ data prep workloads. | |
| Multi-tenancy | ⬤⬤ | ⬤⬤ |
| Rockset supports multi-tenancy — however, it does not recommend scaling to thousands of collections or workspaces | Use a single database instance to serve multiple customers while isolating data and preventing noisy neighbors. |
Simplicity
Quickly build powerful applications
| Capability | Rockset | SingleStore |
|---|---|---|
| HTAP | ⬤⬤⬤ | ⬤⬤⬤ |
| Rockset supports separation of storage and compute | SingleStore's three-tiered "bottomless storage" includes an in-memory rowstore, a disk-based columnstore (persistent cache) and cloud object storage. This separation of storage and compute enables supporting growing data volumes using low-cost storage infrastructure. | |
| Multi-model | ⬤⬤⬤⬤ | ⬤⬤⬤⬤ |
| Rockset supports "compute-compute separation" to allow independent scaling of compute for multiple applications and workloads | SingleStore compute workspaces allow independently scaling compute for different workloads while utilizing a shared database. | |
| Vector search | ⬤⬤ | ⬤⬤⬤⬤ |
| Rockset supports KNN and ANN using FAISS-IVF alone. Indexed vector search in Rockset requires building a vector index for each partition, making it difficult to scale when dealing with changing datasets. The approach is also inefficient because it involves gathering a set of centroids close to the target vector, then applying a predicate to check if each row belongs to those nearby centroids. Indexed vector search is restricted to IVF algorithms and does not extend to more advanced algorithms like HNSW, which offer superior performance and recall. Distance metrics include DOT PRODUCT, COSINE SIMILARITY and EUCLIDEAN DISTANCE. | SingleStore includes both exact K Nearest Neighbor (KNN) and Indexed Approximate Nearest Neighbor (ANN). Supported ANN algorithms include IVF and HNSW along with Product Quantization (PQ) optimizations. Distance metrics include DOT PRODUCT and EUCLIDEAN DISTANCE. | |
| Full-text search | ⬤⬤ | ⬤⬤⬤ |
| Rockset supports full-text and vector search. | SingleStore includes JLucene, the same full-text search engine as used by Solr and ElasticSearch. Its powerful SQL interface enables using filters, aggregations and joins for true hybrid search (vector + full-text search). | |
| Notebooks | ⬤ | ⬤⬤⬤⬤ |
| Rockset does not natively support SQL/Python notebooks, only a SQL editor. It does not include a co-pilot for assistance with writing code. | SingleStore includes native Notebooks that allow developers and data scientists to easily write and collaborate on SQL or Python code | |
| Ecosystem | ⬤ | ⬤⬤⬤ |
| Rockset is built on RocksDB which does not have a large ecosystem of compatible tools. Rockset does offer native connectors for Kafka, DynamoDB, MongoDB®, S3. | With MySQL wire-compatibility, SingleStore benefits from an extremely large MySQL ecosystem. In addition, SingleStore has a large and growing set of native integrations for tools, connectors, frameworks, etc. Learn more. |
Enterprise-ready
Minimize costs, complexity and risks
| Capability | Rockset | SingleStore |
|---|---|---|
| Availability | ⬤⬤ | ⬤⬤⬤⬤ |
| Rockset does not commit to availability SLAs, only support SLAs. | SingleStore provides up to 99.99% uptime SLA and includes support for Point-in-Time Recovery (PITR). This allows customers to run critical applications and workloads and helps mitigate risks on business operations and reputation. | |
| Security | ⬤ | ⬤⬤⬤⬤ |
| Rockset's CMEKs are in private preview. Rockset offers Standard SSO and RBAC. | SingleStore applies encryption to data in transit and data at rest. SingleStore supports Customer Managed Encryption Keys (AWS). SingleStore Helios has secured industry-leading security certifications including ISO 27001 and SOC 2 Type 2. It supports row-level security (RLS), Role Based Access Control (RBAC) and multiple SSO authentication mechanisms including Ping, Okta and OIDC. | |
| Deployment optionality | ⭘ | ⬤⬤⬤⬤ |
| Only available as a cloud service on five AWS regions. | SingleStore can be deployed as a fully managed cloud service or self-managed, either on-premises or on Kubernetes on public cloud infrastructure. It is available as a cloud service in 31 regions across the leading three cloud providers (AWS, Azure, GCP). | |
| Open-source software | ⭘ | ⭘ |
| Rockset is a commercial product | SingleStore is wire-compatible with MySQL and MariaDB. However it is a commercial, enterprise-grade database. It offers a free tier for development use. |
SingleStore beat Rockset
When Adobe chose SingleStore, it evaluated Postgres, ElasticSearch and Rockset, among other products. Learn more about why Adobe chose SingleStore to power their user-facing analytical application, Workfront.
“SingleStore’s latest release is exciting as it pushes what is possible when it comes to database technology, real-time analytics and building modern applications that support AI workloads. We’re looking forward to these new features as more and more of our customers are seeking ways to take full advantage of Gen Al capabilities.”
Watch the SingleStore-Adobe Fireside ChatMatt Newman
Principal Data Architect at Adobe
.svg)
