|
概要（英語）
|
Idle computing resources can reduce storage cost, but node performance is heterogeneous in practice. This paper proposes a performance estimation method for IPFS-based distributed storage that uses secret sharing in heterogeneous environments. The system splits data into shares and stores them on different nodes to improve confidentiality and availability. The study evaluates how heterogeneity affects estimation accuracy. An IPFS Cluster-based prototype is implemented with a custom allocator for share placement. Benchmarks on heterogeneous clusters are used to train Random Forest, Neural Network, and CatBoost models. Results show that higher heterogeneity reduces point-estimation accuracy.
|