Benchmarking Native In-Database TPCx-AI at 100 Terabytes in Ocient Hyperscale Data Warehouse: An Eight-Use-Case Study of Classical and Statistical ML on Relational Primitives

This is the authors' abstract. We don't add key points for this paper.

The growth of enterprise data has created a scalability gap in machine learning (ML) workflows. As data volumes climb into the hundreds of terabytes, extracting raw data to external processing frameworks such as Apache Spark imposes increasing costs in serialization latency, network transfer, and data governance. Audited TPCx-AI results, the standard benchmark for end-to-end ML pipelines, have been limited to scale factor 3,000 (3 TB) to date. This paper presents the architecture and performance of the Ocient Hyperscale Data Warehouse executing eight of the ten TPCx-AI use cases at scale factor 100,000, an aggregate generated dataset size of 100 Terabytes. To our knowledge, this is the largest reported TPCx-AI execution, more than an order of magnitude beyond the largest audited result. Two use cases that require deep-learning model families our engine does not yet implement natively are omitted. For the eight executed use cases, we demonstrate linear or sub-linear scalability across six scale factors (1 GB to 100 TB), meeting the TPCx-AI accuracy target at every scale on every use case except one minority-class classification case whose threshold is also missed by our algorithm-matched Spark implementation, indicating that the gap is intrinsic to the synthetic dataset rather than to either engine. Our goal is to provide an existence proof that a single relational engine can execute classical and statistical ML pipelines at scale on high-end, off-the-shelf enterprise hardware. We detail a database architecture designed to feed query operators directly from NVMe drives to CPU registers via userspace polling. We describe a memory management system using 128 KB fragments within 1 GB hugepages, and how we achieve purely relational execution of iterative algorithms utilizing native matrix data types and operators in standard SQL: no UDFs, no embedded R/Python runtime, and no separate linear-algebra DSL are needed. Furthermore, we discuss a native C++ matrix library and an LLVM-based JIT compiler that translates decision tree models directly into machine code. To position the architectural choices, we report an algorithm-matched comparison against Apache Spark across four scale factors from 100 GB to 100 TB.