Heterogeneous memory management method and device

Abstract

The invention discloses a heterogeneous memory management method and device. The method includes: when a process is created and a memory request is received, determining an applied memory type according to an identifier carried in the memory request; allocating a corresponding memory according to the determined memory type. By the method and device, a system-level process is ensured to be allocated to a fast DRAM (dynamic random access memory) for data processing, and a process for big data processing is reasonably put in an NVM (non-volatile memory) to run. By the method and device, the different types of memories perform data processing of specific types, so that influence on data processing due to access speed difference of the DRAM and the NVM is reduced greatly. On the other hand, when the allocated NVM is recycled, a recycling delay mechanism is adopted, and waiting time for big data processing process access is prolonged, so that the problem of deadlocking possibly caused by access to the DRAM and the NVM is avoided.

Description

technical field [0001] The invention relates to big data processing technology, in particular to a heterogeneous memory management method and device. Background technique [0002] With the rapid development of computer technology, the amount of data has also increased, and the speed of big data processing has become increasingly prominent. In order to ensure its timeliness, the current international academic and industrial circles mainly focus on system software, architecture, and distributed systems. etc. to improve and optimize. in, [0003] For the optimization of system software, technologies such as in-memory database and compiler optimization are mainly proposed to deal with big data processing problems. In-memory databases load related data into memory without incurring the overhead of disk I / O. However, since the memory database provides ACID guarantees, that is, atomicity, consistency, isolation and durability guarantees, applications with weaker consistency requ...

AI Technical Summary

Problems solved by technology

[0006] At present, the industry proposes to use heterogeneous memory (that is, including different types of memory) to process big data. The memory management scheme can perform better when the system memory is small. However, once the system memory reaches TB level, the current The performance of some memory management methods will be greatly reduced. It can be seen that the existing memory management mechanism for heterogeneous memory cannot achieve good management, which further brings about the problem of lower memory application and release efficiency and thus lower system performance.

[0007] For traditional big data processing, data processing is carried out in dynamic random access memory (DRAM), and it needs to be written to disk regularly, which increases overhead

Images