Abstract
dc:descriptionFile system studies are critical to the accurate configuration, design, and continued evolution of storage systems. However, surprisingly little has been published about the use and behaviour of file systems under live deployment. Most published file system studies are years old, or are extremely limited in size and scope. One reason for this deficiency is that studying clusters of storage systems operating at scale requires large data repositories that are difficult to collect, manage, and query. In effect, the demands introduced by performing detailed file system traces creates new and interesting storage challenges in and of itself. This thesis describes a methodology to facilitate the collection, analysis, and publication of large traces of file system activity and structure so that organizations can perform regular analysis of their storage systems and retain that analysis to answer questions about their system's behaviour. To validate this methodology I investigate the use and performance of several large storage deployments. I consider the impact of the observed system usage and behaviour on file system design, and I describe the processes by which the collected traces can be efficiently processed and manipulated. I report on several examples of long standing incorrect assumptions, efficient engineering alternatives, and new opportunities in storage system design.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy - PhD
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Computer Science
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Meyer, Dutch T.
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivs 2.5 Canada
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/54694