Fine-grain dynamic adaptation of the Linux 2.6 virtual memory manager: A first step
The effectiveness of an operating system (OS) depends on how well its management policies suit the needs of workloads running on the system. Most operating systems treat all workloads equally and apply a generalized management policy set at compile time. Since different workloads have different resource usage behaviors, a generalized management policy will suit the needs of some workloads less than others. ^ To compensate for this, many operating systems allow users to tune specific resource management parameters at runtime to suit their particular requirements. Although this adds flexibility to the system, it forces the user to understand the underlying management mechanisms of the OS as well as the resource needs of the workload in question. An alternate solution is to enable the OS to recognize the resource behavior of a workload at runtime and tune its own parameters accordingly, thereby not forcing the user to participate. ^ For an OS to change parameter values at runtime, an understanding of how different parameter values affect the performance of different workloads is needed. This work concentrates on one parameter, in particular, within the Virtual Memory Manager (VMM) of the Linux operating system, SCM or Swap Cluster Max, and analyzes its effect on workload performance. ^ This thesis makes two contributions to the dynamic adaptation of a VMM. First, it describes a methodology that can be used to quantify application performance gains or losses attributable to changing the value of a parameter. Second, this study builds upon the work of Kandiraju , who designated SCM as a potential candidate for dynamic adaptation, by conducting a more thorough study of this parameter, using the described methodology, and reinforcing his results that, indeed, SCM is a good candidate for dynamic adaptation. ^
Portillo, Ricardo Alberto, "Fine-grain dynamic adaptation of the Linux 2.6 virtual memory manager: A first step" (2006). ETD Collection for University of Texas, El Paso. AAI1436513.