Serverless GPU Architecture for Enterprise HR Analytics: A Production-Scale BDaaS Implementation

TL;DR

This work addresses the need for latency-sensitive, compliance-aware analytics in regulated enterprises by proposing a production-oriented BDaaS blueprint that deploys a single-node serverless GPU runtime for TabNet-based tabular inference. The approach delivers substantial performance and cost advantages over multi-node Spark, while incurring only modest security overhead and preserving interpretable, audit-ready explanations through TabNet feature masks. Key contributions include a systematic compliance-aware benchmark, a reproducible Helm-packaged deployment blueprint, and a practical decision framework for CPU, GPU, and Spark trade-offs in regulated settings. The results demonstrate that secure, interpretable, and cost-efficient serverless GPU analytics are feasible and ready for deployment in enterprise and government contexts, with clear guidance on when to adopt this architecture and how to optimize it further.

Abstract

Industrial and government organizations increasingly depend on data-driven analytics for workforce, finance, and regulated decision processes, where timeliness, cost efficiency, and compliance are critical. Distributed frameworks such as Spark and Flink remain effective for massive-scale batch or streaming analytics but introduce coordination complexity and auditing overheads that misalign with moderate-scale, latency-sensitive inference. Meanwhile, cloud providers now offer serverless GPUs, and models such as TabNet enable interpretable tabular ML, motivating new deployment blueprints for regulated environments. In this paper, we present a production-oriented Big Data as a Service (BDaaS) blueprint that integrates a single-node serverless GPU runtime with TabNet. The design leverages GPU acceleration for throughput, serverless elasticity for cost reduction, and feature-mask interpretability for IL4/FIPS compliance. We conduct benchmarks on the HR, Adult, and BLS datasets, comparing our approach against Spark and CPU baselines. Our results show that GPU pipelines achieve up to 4.5x higher throughput, 98x lower latency, and 90% lower cost per 1K inferences compared to Spark baselines, while compliance mechanisms add only ~5.7 ms latency with p99 < 22 ms. Interpretability remains stable under peak load, ensuring reliable auditability. Taken together, these findings provide a compliance-aware benchmark, a reproducible Helm-packaged blueprint, and a decision framework that demonstrate the practicality of secure, interpretable, and cost-efficient serverless GPU analytics for regulated enterprise and government settings.

Figures (4)

• Figure 1: Overall BDaaS system architecture, showing the four main components: (A) Data Pipeline, (B) Serverless GPU Functions, (C) Security Layer, and (D) Monitoring and Observability. The design illustrates the end-to-end data flow from ingestion to inference under IL4/FIPS compliance.
• Figure 2: Throughput comparison across datasets and cluster configurations (steady-state, excluding cold starts). Across all three datasets (HR, UCI Adult, BLS), the single-node GPU serverless pipeline achieves an order-of-magnitude higher throughput than Spark clusters, demonstrating superior scalability particularly for larger batch workloads.
• Figure 3: Latency comparison across different batch sizes and configurations (steady-state measurements). Note the logarithmic scale highlighting the orders-of-magnitude difference between GPU and Spark implementations.
• Figure 4: Cost per 1,000 samples processed (steady-state). Single-node GPU shows 10× cost advantage for larger batches.