Distributed reference counting system and method applied to multi-kernel system

Abstract

The invention relates to a distributed reference count system applied to a multi-kernel system. The distributed reference count system comprises at least two system kernels, an internal storage module, a task manager and a system global reference count module. The distributed reference count system further comprises a plurality of local resource reference modules which correspond to the system kernels one to one. According to the technical scheme designed on the basis of the distributed reference count system, the bottleneck of a traditional operation system reference resource mechanism is overcome, correspondingly, the invention further provides a count method based on the distributed reference count system, the access frequency that the system kernels have access to the global reference count module, the phenomenon that the excessive jitter occurs due to the fact that a certain system kernel monopolizes the system global reference count module is avoided effectively, the service capability of the operation system is improved effectively, and extendibility of the system is improved.

Description

technical field [0001] The invention relates to a distributed reference counting system and a counting method applied to a multi-core system. Background technique [0002] In order to manage various resources, modern operating systems (such as Linux) generally use a reference counting strategy to manage resources, and the reference counting of existing resources is a shared counter. The counters in the kernel can be used for network statistics, recording the usage of various resources, and so on. If multiple processors try to update the same counter, serious race conditions can result. Therefore, traditional operating systems typically use coarse-grained locks to protect shared reference counts in kernel objects. When a process acquires a reference to an object, it first applies for a lock. After the application is successful, the operation counter is incremented by 1, and finally unlocked; Zero, if zero, indicates that no process references the corresponding object, and ...

AI Technical Summary

Problems solved by technology

A simple method is to define the counter as an atomic variable, and use the corresponding atomic addition and subtraction operations. However, the atomic counter uses a complex lock operation, requiring the current CPU to exclusively access the cache of the variable.

A frequently modified atomic counter may cause excessive jitter in the CPU core cache, seriously affecting system performance

[0003] The present invention overcomes the bottleneck of the reference counting mechanism of the traditional operating system, thereby solving the excessive "jitter" caused by the use of coarse-grained locks in the traditional operating system to protect the shared references in the kernel object, which seriously affects system performance issues

Images