Weiya Xi

HiSMRfs: A high performance file system for shingled storage array

HiSMRfs a file system with standard POSIX interface suitable for Shingled Magnetic Recording (SMR) drives, has been designed and developed. HiSMRfs can manage raw SMR drives and support random writes without remapping layer implemented inside SMR drives. To achieve high performance, HiSMRfs separates data and metadata storage, and manages them differently. Metadata is managed using in-memory tree structures and stored in a high performance random write area such as in a SSD. Data writing is done through sequential appending style and store in a SMR drive. HiSMRfs includes a file/object-based RAID module for SMR/HDD arrays. The RAID module computes parity for individual files/objects and guarantees that data and parity writing are 100% in sequential and in full stripe. HiSMRfs is also suitable for a hybrid storage system with conventional HDDs and SSDs. Two prototype systems with HiSMRfs have been developed. The performance has been tested and compared with SMRfs and Flashcache. The experimental tests show that HiSMRfs performs 25% better than SMRfs, and 11% better than Flashcache system.

Simulation for a shingled magnetic recording disk

Shingled magnetic recording (SMR) is a new technology based on partially overlapping, or shingling, adjacent tracks, which is capable of increasing the density of a disk drive up to 10

Concurrent Regeneration Code with Local Reconstruction in Distributed Storage Systems

{\hbox{Tbit/in}}^{2}

Storage controllers and storage control methods

. To write data on overlapping tracks may erase data being written previously. Therefore, without any data translation layer to redirect updates to different locations to avoid unintended data erasures, the SMR disk cannot be used as a conventional disk to perform unrestricted reads/writes directly. Various data translation/mapping strategies have been proposed to allow the disk to perform unrestricted reads/writes. However, these strategies have tradeoffs in terms of system performance and cost, such as disk space utilization. Nevertheless, system performance varies for different applications and under different workload conditions.

Data storage system, method of writing to storage in the data storage system, hard disk and method of forming the hard disk

Reed-Solomon (RS) codes are a standard erasure code choice and their repair cost is so high that it is a penalty for storage efficiency and high reliability. In this paper, we propose a novel class of Concurrent Regeneration codes with Local reconstruction (CRL), that enjoy three advantages: (1) to minimize the network bandwidth for node repair, (2) to minimize the number of accessed nodes, and (3) faster reconstruction in distributed storage systems. We show how they overcome the limitation of RS codes, and we demonstrate that they are optimal on a trade-off between minimum distance and locality. By conducting performance evaluation in both memory and JBOD environments, experimental results demonstrate the performance of the CRL codes.