论文部分内容阅读
新型非易失存储器提供了高访问速度,大容量,字节访问粒度,以及非易失等特性。这些特性将为容错带来新的机遇。本文提出了基于非易失存储器的细粒度检查点。我们在现有操作系统内核的存储管理模块中加入了针对非易失存储器的管理,同时提供了一个非易失堆以供上层应用进行快速的内存分配和对相关对象(或数据结构)的检查点备份。为了实现细粒度检查点,我们将不同的对象(或数据结构)分布在不同的虚拟页面中,并利用硬件提供的页保护机制来实现对数据修改的监控。在此系统中,两个处于不同虚拟页面中的对象(或数据结构)有可能处于同一个物理页面中。对一个对象的修改将不会影响到对另一个对象的监控。以上机制使我们能够对小于一个页面的修改进行监控。和传统的基于页面粒度的检查点工作相比,我们的机制可以有效的减少检查点拷贝的数据量,从而更好的利用存储器带宽,提升性能。
New non-volatile memory provides high access speed, large capacity, byte access granularity, and non-volatile features. These features will bring new opportunities for fault tolerance. This paper presents a non-volatile memory-based fine-grained checkpoint. We have added non-volatile memory management to the existing operating system kernel memory management module and provided a nonvolatile heap for upper-level applications to quickly allocate memory and examine objects (or data structures) Point backup. In order to achieve fine-grained checkpoints, we distribute different objects (or data structures) in different virtual pages and utilize the page protection mechanism provided by the hardware to monitor the data modification. In this system, two objects (or data structures) in different virtual pages may be in the same physical page. A change to an object will not affect the monitoring of another object. The above mechanism allows us to monitor changes to less than one page. Compared with the traditional page-based checkpoint work, our mechanism can effectively reduce the amount of data copied by the checkpoint so as to make better use of memory bandwidth and improve performance.