Abstract:
The I/O bottleneck has become an increasingly daunting challenge for big data analytics along with the explosive growth in data volume. Flash-based SSDs become promising ...Show MoreMetadata
Abstract:
The I/O bottleneck has become an increasingly daunting challenge for big data analytics along with the explosive growth in data volume. Flash-based SSDs become promising to replace the hard disk drives. However, garbage collection (GC) operations in SSDs have a significant impact on the SSD performance, thus leading to performance variability in SSD-based RAIDs. To address this problem, we propose request redirection (RR) by exploiting the asymmetric read-write performance characteristics of SSDs and the hot-spare SSD in SSD-based RAIDs to alleviate the GC-induced performance variability. RR services the incoming read requests to the SSD currently in GC state by reconstructing the read data from other SSDs in the same stripe within SSD-based RAIDs. For the incoming write data to the SSD in the GC state, RR temporarily stores the write data on the hot-spare SSD and concurrently updates the corresponding parity in the SSD-based RAIDs. Extensive evaluations on the RR prototype show that the RR scheme significantly reduces the average response time and alleviates the performance variability, compared with the local GC and global GC schemes.
Published in: IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems ( Volume: 38, Issue: 5, May 2019)