Compression method and device for fault-tolerant nvm persistent process redundant information

Abstract

The invention discloses a method and device for compressing redundant information in a fault-tolerant NVM persistent process. The implementation steps of the invention include allocating write sets, address queues and multiple value queues, and allocating persistent log storage areas in NVM; During the transaction, write all persistent write operation information into the write set; scan the write set, add the corresponding value queue according to the comparison of the old and new values, and add the address queue in order according to the value queue; compress the address information in the address queue The same part of a group of adjacent addresses is eliminated, and the compressed address information is written into the address record; the lower X bits of the N-bit values ​​of all elements in each value queue are written into the value record, so that the redundant information of the value eliminate. The invention can reduce the size of the persistent data, thereby reducing the persistence overhead of the transaction, achieving the purpose of improving the performance of the persistent transaction and prolonging the service life of the NVM.

Description

technical field [0001] The present invention relates to the field of computer persistent memory, in particular to a method and device for compressing redundant information in a fault-tolerant NVM persistent process, which is used to compress redundant information in persistent storage to improve the write performance and lifespan of persistent memory . Background technique [0002] In recent years, with the rapid development of the computer field, the processing capacity of the central processing unit CPU has been greatly improved, while the performance improvement of the storage system is relatively small. There is a huge performance gap between the traditional main memory-auxiliary memory structure. It has become the performance bottleneck of the computer system. [0003] On the one hand, in the traditional main memory-auxiliary memory system, the main memory (usually DRAM, Dynamic Random Access Memory, dynamic random access memory) usually has the characteristics of smal...

AI Technical Summary

Problems solved by technology

from figure 2 As can be seen in the above, no matter what kind of log mechanism, there is a certain overhead in persistence: two write operations (write to the log and write to the persistent heap) must be performed, and each write operation requires data update ( Flush) and operation synchronization (fence) are two time-consuming operations to ensure that the write persistence is completed, and the NVM write operation itself has a long delay

On the other hand, such log persistence only simply writes the address and value information into NVM, and does not deal with the redundancy of information locality during program execution.

Images