Temporal-Spatial Locality in Modern DB Design: From Access Patterns to Time-Series Warehouses

Temporal-spatial locality is no longer abstract—it's guiding how we index, cache, and store data across OLTP and time-series workloads. A locality-focused blog post and Arc, a time-series warehouse, show the pattern in action. ^[1][2]

Temporal-Spatial Locality in DB Design Post 1 argues that locality shapes access patterns, driving indexing, caching, and storage decisions in both transactional and time-series contexts. The takeaway: locality isn’t cute—it’s a design lever that downstream systems must respect. ^[1]

Arc: a high-throughput time-series warehouse Arc conceptually demonstrates locality-grounded design in practice. Here are the core moves:

• Ingest via a binary MessagePack API (fast path) ^[2]
• Stores data as Parquet with hourly partitions for scalable analytics ^[2]
• Queries run on DuckDB for SQL-based analysis ^[2]
• Write throughput clocks ~1.88M records/sec (MessagePack, untuned) ^[2]
• ClickBench on AWS c6a.4xlarge: 35.18 s cold, ~0.81 s hot (43/43 queries) ^[2]
• Core storage uses MinIO as the backend, with S3-style options ^[2]
• Compatible with Line Protocol to ingest from InfluxDB-style tooling ^[2]
• Arc is open-sourced to invite benchmarks, feedback, and collaboration ^[2]
• Designed to store long-term data for InfluxDB, Timescale, or Kafka using Parquet and S3-style backends ^[2]

The pattern? Ingestion speed, columnar storage, and SQL analytics knit together with locality-aware caching and partitioning to support both writes and reads at scale. ^[2]

Closing thought: locality-focused design isn’t theoretical—it’s a practical guide for building scalable OLTP and time-series stacks that you can actually run.

References