agrawal (4)

8/9/2019 agrawal (4)

1/35

Generating Realistic Impressions

for File-System BenchmarkingNitin Agrawal

Andrea C. Arpaci-Dusseau, Remzi H. Arpaci-Dusseau

8/9/2019 agrawal (4)

2/35

For better or for worse,

benchmarks shape a fieldDavid Patterson

2

8/9/2019 agrawal (4)

3/35

8/9/2019 agrawal (4)

4/35

FS images in past: use what is convenient

Randomly generated files ofseveral MB (FAST 08)1000 files in 10 dirs w/ random data (SOSP 03)

188GB and 129GB volumes in Engg dept (OSDI 99)

5-deep tree, 5 subdirs, 10 8KB files in each (FAST 04)

10702 files from /usr/local, size 354MB (SOSP 01)

Typical desktop file system w/ no description (SOSP 05)

1641 files, 109 dirs, 13.4 MB total size (OSDI 02)

8/9/2019 agrawal (4)

5/35

Performance offindoperation

5

Relative

Time

Taken

Disk layout & cache state

File-system logical

organization

8/9/2019 agrawal (4)

6/35

Problem scope

Characteristics of file-system images have strong impact on performance

We need to incorporate representative file-system images in benchmarking & design

6

How to create representativefile-system images?

8/9/2019 agrawal (4)

7/35

Requirements for creating FS images

Access to data on file systems and disk layout Properties of file-system metadata [Satyanarayan81,

Mullender84, Irlam93, Sienknecht94, Douceur99, Agrawal07]Disk fragmentation [Smith97]More such studies in future? A technique to create file-system images that

isRepresentative: given a set of input distributionsControllable: supply additional user constraints Reproducible: control & report internal parametersEasy to use: for widespread adoption and consensus

7

8/9/2019 agrawal (4)

8/35

Introducing Impressions Powerful statistical framework to generate

file-system images

Takes properties of file-system attributes as input

Works out underlying statistical details of the image Mounted on a disk partition for real benchmarkingSatisfies the four design goals

Applying Impressions gives useful insightsWhat is the impact on performance and storage size?How does an application behave on a real FS image?

8

8/9/2019 agrawal (4)

9/35

Outline

Introduction Generating realistic file-system images

Applying Impressions: Desktop search

Conclusion

9

8/9/2019 agrawal (4)

10/35

Overview ofImpressions

10

Impressions

8/9/2019 agrawal (4)

11/35

Properties of file-system metadataFive-year study of file-system metadata [FAST07]

(Agrawal, Bolosky, Douceur, Lorch)

Used as exemplar for metadata properties in Impressions

11

8/9/2019 agrawal (4)

12/35

Features of Impressions Modes of operation for different usages

Basic mode: choose default settings for parametersAdvanced mode: several individually tunable knobs

Thorough statistical machinery ensures accuracyUses parameterized curve fitsAllows arbitrary user constraints

Built-in statistical tests for goodness-of-fit

Generates namespace, metadata, file content, anddisk fragmentation using above techniques

12

8/9/2019 agrawal (4)

13/35

Creating valid metadata Creating file-system namespace

Uses Generative Model proposed earlier [FAST 07]Explains the process of directory tree creationAccurately regenerates distribution of directory

size and of directory depth

13

8/9/2019 agrawal (4)

14/35

Creating namespace

14

Directory tree Monte Carlo runIncorporates dirs by depth

and dirs by subdir counti

Probability of parent selection Count(subdirs)+2

50

60

70

80

90

100

0 2 4 6 8 10 12 14 16Cumulative%o

fdirectories

Count of subdirectories

Directories by Subdirectory Count

DG

Dirs by subdir count

00.020.040.060.08

0.10.120.140.160.18

0 2 4 6 8 10 12 14 16Fraction

ofdirectories

Namespace depth (bin size 1)

Directories by Namespace Depth

DG

Dirs by namespace depth

DatasetGenerated

8/9/2019 agrawal (4)

15/35

Creating valid metadata Creating file-system namespace Creating files: stepwise process

File size, file extension, file depth, parent directoryUses statistical models & analytical approximations

15

8/9/2019 agrawal (4)

16/35

Example: creating realistic file sizes

Pure lognormal distribution no longer good fit Hybrid model: lognormal body, Pareto tail

Fits observed data more accurately, used to recreatefile sizes in Impressions

Con

tribution

tousedspace

Containing file size (bytes, log scale)

LognormalHybrid

16

File Sizes

8/9/2019 agrawal (4)

17/35

Creating files

17

File Size ModelLognormal body, Pareto tail

Captures bimodal curveiS2S4S6S8S9 S1S3S5S7

0

0.02

0.04

0.06

0.08

0.1

0.12

0 8 2K 512K 512M128G

Fractio

n

ofbytes

File Size (bytes, log scale)

Files by Containing Bytes

DG

Files by containing bytes

8/9/2019 agrawal (4)

18/35

Creating files

18

File Extensions Percentile valuesTop 20 extensions account

for 50% of files and bytesiS2S4S6S8S9 S1S3S5S7 E2E4E6E8E9 E3E5E7 E1

0

0.2

0.4

0.6

0.8

1

Desired Generated

Fractio

n

offiles

Top Extensions by Count

cppdllexegifhhtm

jpgnulltxt

others

cppdllexegifhhtm

jpgnulltxt

others

Top extensions by count

8/9/2019 agrawal (4)

19/35

Creating files

19

File DepthPoissonMultiplicative model along

with bytes by depth

00.020.040.06

0.080.1

0.120.140.16

0 2 4 6 8 10 12 14 16

Fraction

offiles


Files by Namespace Depth

DG

Files by namespace depth

16KB

64KB

256KB

768KB2MB

0 2 4 6 8 10 12 14 16Meanby

tesperfile

(log

scale)


Bytes by Namespace Depth

DG

Bytes by namespace depth

iS2S4S6S8S9 S1S3S5S7 E2E4E6E8E9 E3E5E7 E1D2D4D6D8D9 D3D5D7 D1

8/9/2019 agrawal (4)

20/35

Creating files

20

Parent Dir Inverse Polynomial Satisfies distribution of dirs

with file count

0

0.05

0.1

0.15

0.2

0.25

0 2 4 6 8 10 12 14 16

Fraction

offiles


Files by Namespace Depth(with Special Directories)

DG

Files by namespace depthw/ special dirs

i

8/9/2019 agrawal (4)

21/35

Resolving arbitrary constraints

21

8/9/2019 agrawal (4)

22/35

0

0.05

0.1

0.15

8 2K 512K 8M

Fraction

offiles


C

0

0.05

0.1

0.15

8 2K 512K 8M

Fraction

offiles


O

Resolving arbitrary constraints

Accurate both for the sum and the distribution22

Original Constrained

Contrived

for sum

Constraint: Given count of files & size distribution, ensure

sum of file sizes matches a desired total file system size

8/9/2019 agrawal (4)

23/35

Resolving arbitrary constraints Arbitrarily specified on file system parameters Variant of NP-complete Subset Sum Problem

Approximation algorithm based solution (in paper)Oversampling to get additional sample valuesLocal improvement to iteratively converge to the

desired sum by identifying best-fit in current sample While constraints are satisfied, constrained

distribution also retains original characteristics23

8/9/2019 agrawal (4)

24/35

Interpolation and extrapolation Why dont we just use available data values?

Limited to empirical data in input dataset

What-if analysis beyond available dataset

Efficient to maintain compact curve fits and useinterpolation/extrapolation instead of all data

Technique: Piecewise interpolation

24

8/9/2019 agrawal (4)

25/35

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0 2 832

128

512

2K

8K

32K

128K

512K

2M

8M

32M

128M

512M

2G

8G

32G

128G

Fractiono

fbytes

File Size

Piecewise Interpolation

100 GB

50 GB

10 GB

Segment 19Segment 19Segment 19Segment 19

0

0.02

0.04

0.06

0 50 100SegmentValue

File System Size (GB)

Interpola9on: Seg 1

Interpolation technique & accuracyPiecewise Interpolation

25

Each distribution broken down into segments Data points within a segment used for curve fit Combine segment interpolations for new curve

0

0.02

0.04

0.06

0.08

0.1

0.12

8 2K 512K 128M 32G

Fraction

ofbytes

File Size

Extrapolation (125 GB)

RE

File Size extrapolation 125GB

0

0.02

0.04

0.06

0.08

0.1

0.12

8 2K 512K 128M 32G

Fraction

ofbytes

File Size

Interpolation (75 GB)

RI

File Size interpolation 75GBReal

InterpolatedExtrapolated

Real

8/9/2019 agrawal (4)

26/35

File content Files having natural language content

Word-popularity model (heavy tailed)

Word-length frequency model (for the long tail)

Content for other files (mp3, gif, mpeg etc)Impressions generates valid header/footerUses third-party libraries and software

26

8/9/2019 agrawal (4)

27/35

Disk layout and fragmentation

27

8/9/2019 agrawal (4)

28/35

Disk layout and fragmentation Simplistic technique

Layout Scorefor degree of fragmentation [Smith97]

Pairs of file create/delete operations till desiredlayout score is achieved In future more nuanced ways are possible

Out-of-order file writes, writes with long delaysAccess to file-system specific interfaces

FIPMAP in Linux, XFS_IOC_GETBMAP for XFSPerhaps a tool complementary to Impressions

28

File 1 File 2

1 non-contiguous block (out of 8)File Layout Score = 7/8 All blocks contiguousFile Layout Score = 1 (6/6)

8/9/2019 agrawal (4)

29/35

Outline

Introduction Generating realistic file-system images

Applying Impressions: Desktop search

Conclusion

29

8/9/2019 agrawal (4)

30/35

Applying Impressions Case study: desktop search

Google desktop for linux (GDL) and Beagle

Metrics of interest:

Size of index, time to build initial search indexIdentifying application bugs and policies

GDL doesnt index content beyond 10-deep Computing realistic rules of thumb

Overhead of metadata replication?30

8/9/2019 agrawal (4)

31/35

8/9/2019 agrawal (4)

32/35

Impact of metadata and content

Developer aid: understanding impact of different

file system content & different index schemes

32

00.5

11.5

22.5

33.5

Original

TextCache

DisDir

DisFilte

rRelativeIndexSize

Beagle: Index Size

DefaultText

Image

Binary

8/9/2019 agrawal (4)

33/35

Impact of metadata and content

Reproducing identical file-system image to

compare other apps or ones developed later

33

00.5

11.5

22.5

33.5

Origina

l

TextC

ache

DisDir

DisFilte

rRelativeIndexSize

Beagle: Index Size

DefaultText

Image

Binary

Future AppBeagle

8/9/2019 agrawal (4)

34/35

Conclusion Impressions framework for realistic FS images

Representative, controllable, reproducible, easy to use

Includes almost all file system params of interest

Extensible architecturePlug in new statistical constructs, new models for

metadata and content generation Powerful utility for file-system benchmarking

To be contributed publicly (coming soon) http://www.cs.wisc.edu/adsl/Software/Impressions

34

8/9/2019 agrawal (4)

35/35

Nitin Agrawalhttp://www.cs.wisc.edu/~nitina

Impressions download (coming soon)http://www.cs.wisc.edu/adsl/Software/Impressions

35

Questions?

i

Documents

agrawal (4)