| 引用本文: | 孟祥辉,曾学文,陈晓,叶晓舟.FusionCache: 采用闪存的iSCSI存储端融合缓存机制[J].哈尔滨工业大学学报,2017,49(11):66.DOI:10.11918/j.issn.0367-6234.201701022 |
| MENG Xianghui,ZENG Xuewen,CHEN Xiao,YE Xiaozhou.FusionCache: A Fusion Cache Mechanism for iSCSI Target Based on Flash Memory[J].Journal of Harbin Institute of Technology,2017,49(11):66.DOI:10.11918/j.issn.0367-6234.201701022 |
|
| 摘要: |
| 针对原生的iSCSI目标端控制器缺乏独立的缓存模块问题，为了进一步提高存储区域网的整体性能，在iSCSI target软件中引入了一种基于闪存的融合缓存机制FusionCache. FusionCache利用闪存和DRAM组成统一的融合缓存架构，闪存充当DRAM的扩展空间，DRAM分为缓存块元数据区和前端缓存区.元数据区基于基数树管理缓存块元数据，用于加速缓存块的查找；前端缓存区基于回归拟合统计并预测缓存块访问热度，并吸收大量写入对闪存带来的冲击，只允许热点数据进入闪存. FusionCache采用改进的LRU算法对缓存块进行替换，并且在写回过程中考虑iSCSI会话状态.实验结果表明：FusionCache能降低对后端磁盘设备的访问频率，提高I/O响应的速度和吞吐.与只采用DRAM的缓存机制以及原生iSCSI target相比，FusionCache的I/O访问延时分别降低了33%和60%，吞吐分别提高了25%和54%；相较于Facebook提出的Flashcache机制，FusionCache的吞吐性能提高了18%，延时降低了27%；FusionCache还具有良好的读缓存命中率；此外，FusionCache能够减少闪存的写入次数，提高闪存使用寿命. FusionCache提供良好的网络存储效率，并且降低了使用成本.
|
| 关键词: 网络存储性能 缓存机制 iSCSI target 闪存 |
| DOI：10.11918/j.issn.0367-6234.201701022 |
| 分类号:TP393 |
| 文献标识码:A |
| 基金项目:中国科学院战略性先导科技专项课题(XDA06010302),中科院创新研究院前瞻项目(Y555021601) |
|
| FusionCache: A Fusion Cache Mechanism for iSCSI Target Based on Flash Memory |
|
MENG Xianghui1,2,ZENG Xuewen1,CHEN Xiao1,YE Xiaozhou1
|
|
(1.National Network New Media Engineering Research Center, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China; 2.University of Chinese Academy of Sciences, Beijing 100049, China)
|
| Abstract: |
| Focusing on the problem of lack of independent cache module of original iSCSI target controller, we introduce a fusion cache mechanism based on flash memory called FusionCache into the iSCSI target software to further improve the overall performance of the storage area network. FusionCache uses flash memory and DRAM to form a unified fusion cache architecture. The flash memory acts as DRAM's expansion space, and DRAM is divided into cache block metadata area (metadata cache) and front-end buffer area (front cache). The metadata cache manages cache block metadata based on radix tree in order to accelerate the cache block searching; the front cache tallies and predicts the access popularity of the cache block based on regressing fitting model, and absorbs the impact of massive writes on flash memory to ensure that only the hot data is allowed to enter the flash memory. FusionCache uses the improved LRU algorithm to do cache replacement. Besides, it takes iSCSI session's state into account during write-back. The experimental results show that: FusionCache is able to reduce access to backend disk devices, and improve I/O response speed. FusionCache reduces I/O access latency by 33% and 60%, and improves throughput by 25% and 54%, compared to cache mechanism with only DRAM and original iSCSI target, respectively. Compared with Flashcache proposed by Facebook, FusionCache improves throughput by 18% and reduces latency by 27%. FusionCache also has a good read cache hit rate. Besides, FusionCache reduces write amount of flash memory, thus extends its life. FusionCache provides good efficiency of network storage and reduces cost.
|
| Key words: network storage performance cache mechanism iSCSI target flash memory |
