An Empirical Benchmark Dataset for Paillier-Based Privacy-Preserving REST API Gateways

Overview homomorphic_request is a comprehensive benchmark dataset designed to evaluate the performance and reliability of Paillier-based homomorphic encryption in privacy-preserving REST API gateway architectures under realistic operational conditions. The dataset contains detailed telemetry from controlled experiments that simulate production workloads, including concurrency patterns, batch processing, fault injection, and retry mechanisms. Motivation Existing benchmarks for homomorphic encryption systems typically focus on isolated cryptographic operations in controlled environments with minimal concurrency and idealized network conditions. However, real-world deployments face challenges such as variable load patterns, network instability, queuing delays, and cascading failures. This dataset bridges that gap by providing empirical measurements from a realistic system under stress. Dataset Description Experimental Design The dataset captures telemetry from multiple experimental suites targeting different operational aspects: Concurrency scaling: Progressive increases in concurrent request load to identify bottlenecks and throughput limits Batch size variation: Impact of different batch sizes on latency, throughput, and resource utilization Homomorphic operation overhead: Isolated measurements of encryption/decryption and homomorphic computation costs Burst traffic patterns: System behavior under sudden traffic spikes and high-throughput scenarios Fault injection: Controlled introduction of network failures, timeouts, and service degradation with accompanying retry logic Per-Request Telemetry Each request record includes fine-grained timing breakdowns: Client-side encryption latency Full HTTP round-trip time Gateway processing overhead Queue waiting time Homomorphic computation duration Payload sizes (encrypted and plaintext) Success/failure status Failure classification (network, timeout, computation error, validation failure) Retry count and backoff delays Correctness validation: sampled decrypted results compared against expected plaintext computations Metadata and Reproducibility The dataset includes: Run-level metadata: Git commit hash, configuration parameters, hardware specifications, software versions Summary statistics: Per-experiment aggregations including percentile latencies, error rates, throughput metrics Pre-aggregated comparison tables: Side-by-side performance comparisons across different configurations All experiments are fully reproducible using the open-source Go-based benchmarking harness (included), which uses fixed seed values, deterministic scheduling jitter, and reproducible failure injection patterns. Data Format The dataset is available in two formats for broad compatibility: CSV: Easily imported into pandas, R, Excel, or database systems JSONL: Structured format suitable for streaming processing, NoSQL databases, and ML pipelines Applications This dataset has been used for: Tail latency prediction and modeling Failure mode classification and root cause analysis Anomaly detection in encrypted computation workflows Reliability and availability modeling Performance optimization of homomorphic encryption gateways Comparative evaluation of different cryptographic schemes and implementations Target Audience The dataset is intended for researchers and practitioners in: Applied cryptography Privacy-preserving systems and secure computation Systems performance engineering Distributed systems reliability Network security and secure middleboxes Machine learning for systems (performance prediction, anomaly detection) Citation and Availability The dataset and associated benchmarking tools are open-source and publicly available. Researchers using this dataset in publications are encouraged to cite it to enable reproducibility and comparison across studies. Keywords: homomorphic encryption, Paillier cryptosystem, privacy-preserving computation, REST API, performance benchmarking, fault injection, distributed systems, secure computation