Towards self-optimizing memory management
Abstract (Summary)iii Systems today are application driven. Increasing application sizes re-iterate the importance of memory management and increasing application complexity stresses the need for selfmanagement. At the same time, different memory requirements of different applications require that optimizations for memory management be done from a complete system perspective. In the view of this, the goal of this thesis is to take a step towards self-optimizing memory management at all the different levels of the memory hierarchy. This thesis makes three main contributions to the memory management system. First, it undertakes a thorough characterization study for the TLBs and proposes a novel prefetching mechanism that is simple, powerful and adapts to the applications. Second, it presents a dynamic memory allocator that tunes itself to the applications. Finally, towards the goal of developing a self-optimizing VMM, it finds the important VMM parameters that govern the system performance, relates the influence of these parameters to the application/OS characteristics, and provides a solid motivation to set these parameters dynamically.
School Location:USA - Pennsylvania
Source Type:Master's Thesis
Date of Publication: