ABSTRACT:
David Patterson once famously said, "For better or for worse, benchmarks shape a field". Unfortunately, file and storage system benchmarking is in a state of disarray. There is little consensus on workloads that matter and inadequate infrastructure to create realistic benchmarking environments. In addition, more representative benchmarks are often larger and harder to setup, encouraging unrealistic compromises to become common practice.
In this talk I will present my dissertation research which contributes novel techniques and solutions to improve and simplify the task of file and storage benchmarking. In particular, I will present results from a large scale study of file systems that I performed to understand the characteristics of file-system metadata, and talk about three systems that I have designed and built to create realistic benchmarking state (Impressions), representative synthetic workloads (CodeMRI), and to enable scalable storage benchmarking (Compressions), respectively. In addition, I will briefly touch upon research that I have conducted in other areas such as solid-state storage devices, reliability issues in the storage stack, and deconstructing storage clusters like EMC Centera.
BIO:
Nitin Agrawal is a sixth year graduate student at the University of Wisconsin-Madison working with Professors Andrea Arpaci-Dusseau and Remzi Arpaci-Dusseau. His research has focused on various aspects of file and storage system design, evaluation, and implementation. During his graduate studies he has interned at Microsoft Research Redmond, Microsoft Research Silicon Valley, and IBM Almaden Research. His work has received a Best Paper Award at FAST 2009 and a top paper selection at FAST 2007. More details about his research are available at http://www.cs.wisc.edu/~nitina
# # #