Unikernel-Based Dynamic Memory Tuning Study

Virtualization is one of the key components of cloud computing technology.Mainstream virtualization technologies include virtual machines,containers,and Unikernel.Compared with the former two technologies,Unikernel has both high isolation and portability,and has high practical value.One of the problems with Unikernel is that it is not debuggable.That is,if the memory is not properly allocated at runtime,it will affect the system performance.At runtime,its memory allocation cannot be changed.For a typical application environment with this problem-running JVM(JAVA virtual machine),a system that dynamically adjusts the memory allocation of the system(Unikernel)and application(JVM)at runtime is designed and implemented.Specifically,a JVM balloon is implemented,that is,when the system has insufficient memory,the memory is borrowed from the JVM by calling the JVM memory allocation interface,and is mapped to the system memory.When the system memory is idle(not frequently used),the system is released.This memory space.At the same time,this paper combines ant colony algorithm to implement an efficient prediction-based dynamic memory adjustment strategy,which can allocate memory reasonably through JVM balloon technology.This achieves the goal of dynamically allocating memory between the application and the system under runtime conditions.The system prototype was implemented based on the typical Unikernel system OSv.Analysis of the test results shows that the method can effectively change the JVM heap size and accurately detect and quickly process page faults.After adopting this technology,the system performance can be greatly improved under the condition of stabilizing a small amount of I/O.For example,when the JVM memory occupancy rate is 80%,a small amount of data is written to the system multiple times,and the running time can be increased by up to 40%;In burst I/O conditions,system performance is not affected by this technology.