Storage Codes and Data Retrieval
summary
Compares queuing models for coded storage and measures download latency using simulations and analysis: a 10K fork-join simulation with K=1,5,10 shows spreading pieces across many disks reduces tail latency even versus full 10× replication, and doubling storage can approach replication performance; derives upper bounds via Kth order statistics of exponentials and gives closed forms involving beta functions; analyzes availability codes with two disjoint repair groups and shows MDS performance scales as K out of N (effective service K/N), while replication and availability codes trade latency versus storage and fault‑tolerance (example: 15→30 gives ≈½ mean time, some availability codes ≈2/3).
AI-written summary · about