Data Delta Based Hybrid Writes for Erasure-Coded Storage Systems

Erasure coding is widely used in storage systems since it can offer higher reliability at lower redundancy than data replication. However, erasure-coded storage systems have to perform a partial write to an entire erasure coding group for a small write, which causes a time-consuming write-after-read. This paper presents an efficient data delta based hybrid write scheme, named DABRI, which supports fast partial writes for erasure-coded storage systems. DABRI uses data deltas that are the differences between latest data values and original data values to bypass the computation of parity deltas and the read of old data. For a series of n partial writes to the same data, DABRI performs log-based data and parity updates for the first write, and takes in-place data updates and log-based parity updates for the last n- 1 writes. This enables new data to be written into data nodes and parity nodes in parallel, and the overhead of data reads and parity updates can be mitigated. Based on DABRI, we design and implement an erasure-coded prototype storage system that can deliver high-performance for small-write-intensive applications. Experimental results on real-world traces show that DABRI can successfully improve the I/O throughput by up to 77.41%, compared with the state-of-the-art cited in this paper.