A Dell white paper evaluating the Regatta database running on PowerFlex systems found it was a transformative system addressing the inherent limitations of traditional databases, such as sharding, complexity, costs, performance, and scale.

Regatta produces its eponymous, scale-out, transactional (OLTP), analytic (OLAP) relational (OLxP) and vector database, with extensibility to semi-structured and unstructured data. It’s a drop-in replacement for Postgres and can scale to tens of thousands of nodes.

PowerFlex is Dell’s hyperconverged infrastructure (HCI) offering, an alternative to the vSphere-only VxRail, supporting various hypervisors, including VMware, and Kubernetes. It comes with software-defined, scale-out, block storage, and also supports file services. There is a single global namespace for enhanced capacity, unified storage pool management, and it can be deployed in disaggregated (independent compute and storage layers) architectures. Lengthy computations are performed in a massively parallel and distributed way across a potentially large number of nodes, resulting in meaningfully faster execution. PowerFlex SW storage services can be deployed in AWS and Azure.

The 42-page “Regatta Database Performance and Best Practices on Dell PowerFlex” white paper details transactional, analytical and streaming Kafka data performance tests. The transactional test demonstrated a total physical read and write of nearly one million IOPS.** **Once all eight nodes in the test config completed a short transactional ramp-up phase, there was steady performance in terms of IOPS and bandwidth.

In the analytics workload test, out of the nine complex distributed queries that were tested, five compute-bound ones demonstrated linear scalability as the cluster size increased. Four other storage-bound queries did not show additional scalability across cluster expansion. The 4-node Regatta cluster configuration fully saturated the underlying available PowerFlex storage, hence adding compute nodes alone did not yield further performance gains.

Because PowerFlex supports dynamic expansion and linear scale, doubling the number of PowerFlex storage nodes would result in a corresponding doubling of the bandwidth.

The Kafka streaming data test showed successful ingestion of one billion rows into Regatta, with 632,520,591 messages consumed by every 5-second interval. The row insertion rate into the Regatta table exceeded 1 million rows per second with a single Kafka client. There was an average write latency of 0.15 ms per PowerFlex volume during data inserts into the Regatta database table, and an average write bandwidth of 138 MB/s. Using multiple Kafka clients could potentially increase the load injected into the Kafka Connect cluster, thereby improving performance on the database side as well.

The white paper concludes: “The combination of Regatta’s high-performance distributed database capabilities with Dell PowerFlex high-performance storage infrastructure offers a transformative solution. This combined approach addresses the inherent limitations of traditional databases (including sharding, complexity, costs, performance, and scale) and leverages the strengths of both technologies to deliver robust, scalable, and efficient database architecture for both high-performance transactional and analytics/AI workloads.”