Track-based Translation Layers for Interlaced Magnetic Recording
summary
Evaluation shows read-modify-write incurs up to 2.5× write amplification and high latencies on VM block traces, while three device-side track translation layers—track flipping, selective track caching (100-track cache in experiments) and dynamic track mapping—cut amplification to about 1.1× and bring latency close to CMR for many workloads. Implementations need only a few megabytes of metadata (≈2.5 MB track map for a 20 TB drive) and run periodic, bounded reorganization; simulations validate practicality.
AI-written summary · about