• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Symposium on Mass Storage Systems and Technologies >Fast and failure-consistent updates of application data in non-volatile main memory file system
【24h】

Fast and failure-consistent updates of application data in non-volatile main memory file system

机译：非易失性主存储器文件系统中应用程序数据的快速且一致的更新

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Modern applications have their own update protocols to remain failure consistency. However, these protocols are implemented without a comprehensive understanding of the persistence properties of the underlying file systems and typically optimized for disk-based storage. As a result, they are complex, error-prone, and exhibit disappointing performance on emerging fast non-volatile memories (NVMs) due to excessive data copies. In this paper, we propose a novel file system optimized for nonvolatile main memories (NVMMs), FCFS, that offers a series of easy-to-use file-based interfaces to enable both correctness and high performance for applications to consistently update their data on NVMM storage. Specifically, FCFS enables applications to atomically and selectively update multiple files efficiently on NVMM storage based on their consistency semantics. To this end, FCFS employs an NVMM-optimized write-ahead logging (NoW) mechanism to reduce the consistency cost by fully leveraging NVMM's byte addressability and high concurrency properties but without relying on the page-cache layer, which consists of two key techniques. First, Hybrid Fine-grained Logging decouples file system metadata log and data log to avoid the false sharing of log data and achieve a good tradeoff between the copy cost and log tracking overhead. Second, Concurrently Selective Checkpointing allows asynchronous checkpointing to be performed in parallel and with minimum data copies in order to improve its efficiency. Evaluations on an emulated NVMM device demonstrate that FCFS's failure-consistent update protocol outperforms conventional protocols (i.e., write-ahead logging and shadow paging) by up to 276% and FCFS-based applications outperform the original ones by up to 93%. Importantly, FCFS requires modifications to no more than 0.06% of the original code for each ported application in order to support failure consistency.
机译：现代应用程序具有自己的更新协议，以保持故障一致性。但是，这些协议是在没有全面了解底层文件系统的持久性属性的情况下实现的，并且通常针对基于磁盘的存储进行了优化。结果，它们是复杂的，容易出错的，并且由于过多的数据复制而在新兴的快速非易失性存储器（NVM）上表现出令人失望的性能。在本文中，我们提出了一种针对非易失性主存储器（NVMM）进行了优化的新颖文件系统FCFS，它提供了一系列易于使用的基于文件的界面，以使应用程序能够正确且高性能地持续更新其数据。 NVMM存储。具体来说，FCFS使应用程序能够根据其一致性语义在NVMM存储上有效地原子地和选择性地更新多个文件。为此，FCFS采用了NVMM优化的预写日志记录（NoW）机制，通过充分利用NVMM的字节寻址能力和高并发属性来降低一致性成本，而不依赖于由两种关键技术组成的页面缓存层。首先，混合细粒度日志将文件系统元数据日志和数据日志解耦，以避免错误共享日志数据，并在复制成本和日志跟踪开销之间取得良好的平衡。其次，并发选择检查点允许异步检查点以最少的数据副本并行执行，以提高其效率。在仿真的NVMM设备上进行的评估表明，FCFS的故障一致更新协议比常规协议（即预写日志记录和影子分页）的性能高出276％，而基于FCFS的应用程序的性能则比原始协议高93％。重要的是，FCFS要求对每个移植的应用程序进行的修改不超过原始代码的0.06％，以支持故障一致性。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号