Erasure Coding

Erasure coding stores data as coded fragments so that a subset can reconstruct the original object, improving capacity efficiency relative to full replication while introducing encoding, repair, placement, and tail-latency trade-offs.

核心思想

An (k, m)-style layout splits data into k data fragments and m parity fragments; sufficient surviving fragments reconstruct lost data. System behavior depends on code choice, fragment size, placement across failure domains, network bandwidth, CPU/accelerator encoding cost, and repair scheduling.

为什么重要

Erasure coding is a storage-system mechanism, not just a capacity ratio. It changes write amplification, small-write handling, degraded-read latency, and background repair traffic. Results must state fault model, layout, workload, and whether repairs share foreground resources.

关键观察 / 隐含假设

• 观察：code geometry and placement interact with storage management. DisCoGC-FAST26 accounts for stripe alignment when reclaiming stale ranges.
• 观察：repair and read paths can dominate in real deployments. DRBoost-FAST26 and McQueen-FAST26 study storage-system mechanisms under reliability constraints.
• 假设：capacity savings outweigh repair and tail costs. TapeOBS-FAST26 and LESS-FAST26 show that media, workload, and observability boundaries matter.

设计空间与取舍

• Replication vs coding：coding saves space while increasing computation and multi-fragment coordination.
• Small writes vs stripe efficiency：partial-stripe updates can add read-modify-write or buffering cost.
• Repair bandwidth vs foreground SLO：faster repair can contend with client traffic.

引用本概念的论文

• DRBoost-FAST26 — reliability/coding storage mechanism.
• McQueen-FAST26 — coded storage evaluation context.
• DisCoGC-FAST26 — stripe-aware reclamation.
• TapeOBS-FAST26 — media/reliability observability boundary.
• LESS-FAST26 — storage-system trade-offs.