Multigranularity Space Management Scheme for Accelerating the Write Performance of In-Memory File Systems
| Autor: | Ting Wu, Kai Liu, Bingyi Liu, Qingfeng Zhuge, Edwin H.-M. Sha, Chunhua Xiao |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Hardware_MEMORYSTRUCTURES
Memory hierarchy Computer Networks and Communications Computer science Append Linux kernel Parallel computing Computer Science Applications Phase-change memory Non-volatile memory Memory management unit Control and Systems Engineering Hardware acceleration Central processing unit Electrical and Electronic Engineering Information Systems |
| Zdroj: | IEEE Systems Journal. 14:5429-5440 |
| ISSN: | 2373-7816 1932-8184 |
| DOI: | 10.1109/jsyst.2020.2975673 |
| Popis: | Emerging nonvolatile memory (NVM) techniques, such as phase change memory (PCM), spin-transfer torque magnetic random access memory (STT-MRAM), and resistive random-access memory, are promising for high-performance data process by reserving data in the memory hierarchy. Many persistent memory file systems are tailored to achieve high performance by exploring the advanced features of the NVM and the hardware memory management unit (MMU) in the CPU. However, with the efficient storage device and the hardware acceleration, the write routines in persistent memory file systems pose considerable overhead since repeatedly allocating free blocks and constructing the file mapping structure are time consuming. In this article, we propose a new multigranularity space management scheme (MSMS) to accelerate the write performance. The MSMS employs multigranularity structured blocks whose mapping structure is proactively constructed to slash the overhead of allocating new space and constructing the file mapping structure. Moreover, we present efficiently dedicated space allocation algorithms for different write modes. For append write, we present a file-size- and buffer-size-based allocation (FBA) algorithm to efficiently allocate the appropriate blocks. And for copy-on-write, we present an updating data and offset-based allocation algorithm to preferentially allocate structured huge blocks for reducing the overhead of invoking allocation routines. Based on the new design, we have implemented the MSMS for SIMFS in the Linux kernel. Experimental results show that the MSMS significantly reduces the times of invoking allocation routines. The average append write and copy-on-write performance with the MSMS improve by 16.34 $\%$ and 7.51 $\%$ , respectively. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|