Enhancing system performance using persistent ram modules as storage class memory / Tebra A Moussa Jummah

• Main Author: Tebra A Moussa , Jummah
• Format: Thesis
• Published: 2022
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://studentsrepo.um.edu.my/14430/1/Tebra_A_Moussa.pdf; http://studentsrepo.um.edu.my/14430/2/Tebra_A_Moussa.pdf; http://studentsrepo.um.edu.my/14430/

Throughput and time latency are critical performance metrics of most application systems; thus, any underlying storage technology must provide the best of both metrics. Since hard disk drive (HDD) and solid-state drive (SSD) became an I/O bottleneck performance for most intensive data applications, a new storage technology must be produced to address this issue. Storage class memory (SCM) emerged as the new promising technology with byte-addressable, high access time, and persistent features. In this research, an SCM emulator was implemented using a kernel module based on a RAM named ZRAM, which served as a general-purpose RAMDISK feature with persistence. Further, a test has been conducted on the implemented emulator with persistent RAMDISK and PMEM. The experiments conducted in this research were done through three stages: the first of which involved testing the workload within different data placement devices; the second stage involved running of tests upon a collection of disk filesystems, RAM filesystem, and persistent memory filesystem (PMFS). At the third stage, experiments were conducted to examine the effect of moving data files on performance. The implemented emulator persistent ZRAM (PZRAM) achieved superior performance as compared to HDD and SSD with a performance improvement of 14290% and 1167% respectively, a slide higher performance than PMEM with 2.3% improvement and almost similar performance of persistent RAMDISK. Additionally, the proposed PZRAM with TMPFS running on top of it has provided better performance with 11.83% over PZRAM with ext4. Further, this research provided comparative experiments on the effect of filesystem and moving data files on throughput and latency performance.