A Full-system Simulation Platform For Computation Offloading In Mobile Edge Computing

Mobile devices have become an integral part of people`s lives.Although the development of mobile device hardware is fast,it cannot still meet the increasing needs of performance and low power.To address the above issue,an approach is used widely to handle compute-intensive workloads of mobile devices on high-performance edge cloud servers.In the current study of computation offloading in mobile edge computing,the research platforms are numerical analysis software or hardware platform.The former is controllable but unobservable;while the latter is observable but uncontrollable.Therefore,a kind of platform that is both controllable and observable is lacking.Due to the issue,we design a simulation platform named edge-gem5.Edge-gem5 is a distributed,full-system simulation platform with a computation offloading framework.It allows the researcher to customize the hardware parameters of the nodes in the system and the network parameters of the system.The design of edge-gem5 is to transplant the computation offloading framework RAPID to the clock-accurate distributed full-system simulation platform dist-gem5.During the transplant process,we study and customize dist-gem5 to simulate heterogeneous distributed system.Meanwhile we deeply study the computation offloading mechanism,address the technical conflict between RAPID and dist-gem5,and finally realizes the transplantation.Finally,this thesis provides an easyto-use method to make program offloadable,and use it to make two programs Sudoku and VirusScanning offloable.Finally,we test the design of the simulation platform,and Experiments show that edge-gem5 can simulate a distributed system with a computation offloading framework.The framework is two-level including a client and an edge computing server.Moreover,it provides 16 indexes of system performance.In addition,the thesis explores the impact of system parameters(CPU frequency,L1 cache size of mobile devices and network speed)on system performance and offloading strategy.The experiment shows that,when the server hardware is constant,the higher CPU frequency of the mobile device can make client more powerful so that the offloading decision prefers to execute three workloads(NQueens,Sudoku and VirusScanning)locally.Only when running VirusScanning,the network speed and mobile device's L1 cache size can enhance system performance,so that making offloading policy tend to execute the program locally and remotely separately.