Abstract
A large area of research for the past several decades has been dedicated to modeling processor caches in order to estimate cache miss rates for workloads. These models were usually accurate, but were not designed to be efficient. My thesis work involved estimating cache miss rates inside an operating system scheduler, where accuracy was not critical, but estimates had to be computed efficiently at runtime. In my talk, I will give a short overview of the existing cache modeling techniques, describe the challenges in creating a high-performance cache model, and present the new online models that I developed.
Biography
Alexandra Fedorova is a Ph.D. student at Harvard University. She is a part of the systems research group. In the past she worked on high-performance distributed file systems. Her current focus is on enhancing operating systems for achieving better synergism between next-generation processors and applications. For her thesis work, she has studied performance of chip-multithreaded processors and developed a new operating system scheduling algorithm that resulted in improved utilization of hardware resources and better performance. Alexandra has close relationship with Sun Microsystems Research labs, where she has been an intern for 2.5 years. She will be graduating in the Spring of 2006.