Tape Storage Technology: Enterprise Deep Archive and the Berg Cold‑Data Engine

This article revisits magnetic tape, once the dominant medium for music distribution, and explains why it remains vital for cold‑data storage in modern enterprises. While consumer tape has disappeared, tape’s low cost, long lifespan, and high capacity make it a key component of deep‑archive solutions.

The hardware stack consists of tape libraries (also called tape cabinets), drives (tape machines), robotic arms, and slot trays. Libraries come in two forms: standalone cabinets with 400‑900 slots (≈5‑10 PB) and rack‑mounted cabinets that can hold thousands of slots (hundreds of PB). The main tape media are LTO (an open standard) and IBM’s 3592, with current generations offering 12‑20 TB per cartridge and future LTO‑9 promising 18 TB.

Compared with HDDs, tape offers dramatically lower total cost of ownership, superior reliability, and far greater theoretical density growth. However, tape is a sequential‑access device: write throughput is high (300‑400 MB/s) but read latency can reach minutes due to mechanical seeking, making it best suited for large, infrequently accessed files.

Cloud Object Storage (COS) provides a Deep Archive tier that abstracts tape hardware behind a simple object‑storage API. It targets use cases such as long‑term log retention, archival video, and compliance data, offering ultra‑low storage fees while handling data ingestion, retrieval, and lifecycle management.

Tencent’s Berg engine is a purpose‑built cold‑data storage system that runs on top of tape. Its architecture includes three modules: Captain (task scheduler with persistence), IceWorker (executes tasks and talks to the tape library), and IceCenter (resource manager). Key designs include real‑time clustering of objects to improve read locality, fault‑tolerant handling of hardware failures, and support for arbitrary erasure‑coding schemes.

Berg defines clear data‑flow processes: ingestion (data sinking), retrieval (data warming), deletion (three‑stage deletion to reclaim tape space), and repair (handling tape failures with EC reconstruction). Security is ensured through end‑to‑end CRC checks, periodic offline verification, and multi‑level metadata snapshots.

Overall, tape storage is ideal for genuine cold‑data workloads that tolerate hours‑to‑days latency and require cost‑effective, durable archiving. The article concludes that while tape offers immense potential, its deployment demands careful operational planning, and cloud‑based deep‑archive services remain a practical entry point for most customers.