Add Type Awareness to File Systems Daniel Peek Jason Flinn Facebook - PowerPoint PPT Presentation

TrapperKeeper: The Case for Using Virtualization to Add Type Awareness to File Systems Daniel Peek Jason Flinn Facebook University of Michgan

Trapper Keeper • Need access to type-specific metadata – Searching, Organization, Presentation • Extracting metadata is hard – Lots of file types out there – Custom code required for each type • A better way to get metadata University of Michigan 2

The Plug-in Solution • Developers make plug-ins for each type Metadata-Using Application Query: MP3s with Composer = Mozart Metadata Engine JPEG Plug-in MP3 Plug-in … University of Michigan 3

The Plug-in Solution • Lots of work for developers Mac OS X Spotlight Metadata Preview University of Michigan 4

The Long Tail • How big is this problem? • Big (Agrawal et al.) • Uneconomical to support all types University of Michigan 5

The TrapperKeeper Solution • Already have apps that parse these files – Apps expose information through GUI Date Time Original = 2007:11:22 19:21:14 University of Michigan 6

The TrapperKeeper Process • Once Per Application – Trap the application • Once Per File – Use trapped application to parse the file – Capture displayed output metadata preview file … University of Michigan 7

Trapping Applications • Run app inside a VM – Contains app effects • Make app open a dummy file • Snapshot at moment of open() – About to execute file parsing behavior Dummy University of Michigan 8

Parsing with Trapped Apps • Restart VM • Switch files File To Dummy Parse University of Michigan 9

Accessibility Window Tab Pane … Text Label: “Date Time Original” Text: “2007:11:22 19:21:14” … University of Michigan 10

Accessibility Window Tab Pane … Text Label: “Date Time Original” Text: “2007:11:22 19:21:14” … University of Michigan 11

Guided Extraction University of Michigan 12

Guided Extraction University of Michigan 13

Guided Extraction University of Michigan 14

Execute Features • Snapshot window in VM metadata Metadata System Preview University of Michigan 15

TrapperKeeper Results • Makes it easy to extract metadata – No development skill – No source code – Just be able to use the application • Successful use – All GUI apps in Ubuntu 7.10 in a day – Parses over 100 file types – Rate of 318 files/hour University of Michigan 16

Tricky Situations • Application has no accessibility support • Application does not expose metadata • Application needs external info to parse – Configuration files – License servers – Internet connections University of Michigan 17

Tricky Situations • Performance: a sudden influx of files – Fresh installation – Download from digital camera • Which metadata is the right metadata? University of Michigan 18

Questions University of Michigan 19