Title: Coupling Decentralized Key-Value Stores with Erasure Coding
Conference (SoCC’19): Link
Journal (): Link
When consistent hashing is combined with erasure coding and applied to decentralized KV-Store, there exists high parity updates and cannot handle degraded read caused by scaling. This paper introduces introduces an erasure coding model FragEC, which incur no parity update for storage scaling for decentralized KV-Store. It addresses the degraded read issue by combining consistent hashing. It then implements ECHash, an in-memory KVStore based on memcached. Experiment shows that ECHash achieves better scaling performance than baseline erasure coding implementation over memcached.
FragEC, or fragmented erasure coding: one data chunk can be divided into sub-chunks and can mapped to multiple nodes (for consistent hashing with multiple rings).
Design consistent hashing based on multiple hash rings, which partition nodes into multiple isolated zones. One hash ring corresponds to one zone.
Improve node repair performance over baseline approach by supporting the repair of sub-chunks.