Google Cloud Bigtable vs RonDB

Google Cloud Bigtable is a managed, low-latency NoSQL database for machine learning, operational analytics, and user-facing applications. It stores structured, semi-structured, and unstructured data in key-value and wide-column formats. SQL querying includes an API with more than 200 functions and support for aggregations, windowing, geospatial and JSON processing, and vector search. Compute scales horizontally, and instances can span up to eight regions with automatic replication between clusters. Google says multi-region instances offer 99.999% availability and protection against a regional failure. Integrations listed include Apache Spark, Hadoop, GKE, Dataflow, Dataproc, Vertex AI Vector Search, BigQuery, and LangChain. Bigtable provides Apache HBase and Cassandra APIs, plus migration tools for HBase, Cassandra, DynamoDB, and Aerospike. Security controls include IAM, customer-managed encryption keys, audit logging, and table-, column-, or row-level access control. Deployment is cloud-based. Google describes its capacity-based low-latency compute as more favorable for larger workloads than small applications. A 10-day trial is available subject to eligibility requirements.

RonDB is an open-source distributed storage engine for real-time applications and a fork of MySQL NDB Cluster. Its shared-nothing design supports up to 1 PiB of storage, with online horizontal and vertical scaling for storage and MySQL-server compute. Applications can use the MySQL Server API for SQL-compatible access and parallel queries. ACID transactions are provided through MySQL-server support; in-memory data is persisted using REDO logs and checkpoints.

RonDB describes its performance aims as low latency, high availability, high throughput, and scalable storage. It supports AI application storage integrated with Spark and Kafka, while Rondis implements a Redis server using RonDB storage and the Redis wire protocol. Deployment options include local testing, Docker, cloud deployments using your own AWS, Azure, or GCP credentials, and on-premises deployment by inquiry. Replication can be synchronous active-active across availability zones or asynchronous across regions with conflict resolution. RonDB is used as the online feature-store database for machine-learning applications in Hopsworks. The serverless plan through Hopsworks is free, with quotas on Feature Groups and memory usage.