Drive-Thru: Drive-Thru: Fast, Accurate Evaluation of Fast, - - PowerPoint PPT Presentation

1

Drive-Thru: Drive-Thru: Fast, Accurate Evaluation of Fast, Accurate Evaluation of Storage Power Management Storage Power Management

Daniel Peek Jason Flinn

University of Michigan

2

Power Management Needed Power Management Needed

• Battery lifetime is limited
• Storage is an energy hog
• Power management effective
• Many competing techniques

[Helmbold 96], [Weissel 02], [Douglis 94], [Papthanasiou 03]

How can we evaluate possible power management strategies?

3

Design Goals Design Goals

Storage power management evaluation should be: Fast

– Want to explore many policies/parameters/workloads

Accurate

– Both time and energy predictions are important

Portable

– Easy to apply technique to other file systems/OSs

4

Trace Replay Trace Replay

Prerecorded Trace Trace Replay Tool Write (file1) Stat (file2) Delay 5s Fast Accurate Portable

5

Write (file1) Stat (file2) Delay 5s

Trace Replay Without Idle Time Trace Replay Without Idle Time

Prerecorded Trace Fast Accurate Portable Trace Replay Tool

6

Simulation Simulation

Simulator Prerecorded Trace Write (file1) Stat (file2) Delay 5s Fast Accurate Portable

?

Estimated Time & Energy Trace Replay Tool

7

Outline Outline

• Motivation
• Design
• Validation
• Case Study
• Related Work
• Conclusion

8

The Big Idea The Big Idea

• Time Dependent

– Disk spindown – Writeback of dirty blocks in the buffer cache

• Time Independent

– Mapping reads to disk blocks – Satisfying accesses from the buffer cache

9

Layers of Power Management Layers of Power Management

Time-Dependent Activity Time-Independent Activity Satisfying Accesses File System Buffer Cache Disk Writeback of dirty blocks Spindown Mapping reads to disk blocks Send modifications to network server Accesses

10

File System Log Device Log Drive-Thru Replay Tool

A Hybrid Methodology A Hybrid Methodology

The accuracy of trace replay without the cost

• Replay time independent behavior without idle time
• Simulate time dependent behavior

Base Trace Prerecorded Trace Write (file1) Stat (file2) Delay 5s Simulator

Estimated

Time & Energy

11

Base Traces Base Traces

File System Log Device Log Base Trace

Write (file1) Write sector 8 Write (file1) Write sector 8 Stat (file2) Stat (file2) Stat (file2) Read sector 4 Stat (file2) Read sector 4 Sync Sync

12

Merging Replay and Simulation Merging Replay and Simulation

Time & Energy Drive-Thru Simulator I/O Simulator

Write File1 Device Write Stat File2 Device Read Delay

• Fast: All idle time is simulated
• Portable: Drive-Thru < 1000 lines

13

Drive-Thru Simulator

Drive-Thru Operations Drive-Thru Operations

• Delay:
• Coalesce:
• Reorder:

Drive-Thru Simulator Drive-Thru Simulator

14

Outline Outline

• Motivation
• Design
• Validation
• Case Study
• Related Work
• Conclusion

15

Validation Setup Validation Setup

How accurate is Drive-Thru?

– Compare to most accurate method, replay with idle times

• Ran six 45-minute trace segments

– iPAQ 3870 – Ext2 file system – Hitachi 1 GB Microdrive

16

Validating with ext2 Validating with ext2

• Drive-Thru prediction on average within 0.21%

1000 2000 3000 4000 5000 Purcell Berlioz Messiaen NFS15 Energy (J) Trace Replay with Idle Times Drive-Thru

17

1000 2000 3000 4000 5000 Purcell Berlioz Messiaen NFS15 Energy (J) Trace Replay with Idle Times Drive-Thru

Validating with ext2 Validating with ext2

• Drive-Thru prediction on average within 0.21%

18

Validating with ext2 Validating with ext2

• Drive-Thru prediction on average within 3%

100 200 300 400 Purcell Berlioz Messiaen NFS15 File System Energy (J) Trace Replay with Idle Times Drive-Thru

19

Network File System Validation Network File System Validation

How Accurate is Drive-Thru for network?

– Compare to trace replay with idle times – Compare over 3 power management modes

• Blue file system [Nightingale 04]
• 802.11b card

20

802.11b Power Management 802.11b Power Management

802.11b power management modes:

• Continuously Aware Mode (CAM)

– High performance, high power

• Power-Saving Mode (PSM)

– Low performance, low power

• Self-Tuning Power Mangement (STPM) [Anand 03]

– Adaptively toggles between CAM and PSM

21

• Drive-Thru prediction on average within 7%

Validating Network Predictions Validating Network Predictions

1000 2000 3000 4000 Purcell (CAM) Purcell (PSM) Purcell (STPM) NFS15 (CAM) NFS15 (PSM) NFS15 (STPM) File System Energy (J) Trace Replay with Idle Times Drive-Thru

22

Outline Outline

• Motivation
• Design
• Validation
• Case Study
• Related Work
• Conclusion

23

Case Study: Local Storage Case Study: Local Storage

Proposed power management strategies:

• Flush on write [Papathanasiou 02, Weissel 02]

– When a dirty block is written, writeback all dirty blocks

• Flush on spindown [Weissel 02]

– Before the disk spins down, writeback all dirty blocks

• Increase writeback delay [Papathanasiou 03, Weissel 02]

– Increase opportunities for write aggregation

Ran 6 traces on ext2 with Drive-Thru (8-28 hours each)

24

Case Study: Local Storage Case Study: Local Storage

50 100 150 200 250 10 20 30 40 50 60 Writeback Delay (s) File System Energy (J)

Default Flush on Write Flush on Spindown Flush on Both

• Complete evaluation over 40,000x faster than trace replay

25

50 100 150 200 250 10 20 30 40 50 60 Writeback Delay (s) File System Energy (J)

Default Flush on Write Flush on Spindown Flush on Both

The 2-Second Peak The 2-Second Peak

Buffer Cache Trace Replay Tool Device Time (s) 0 1 2 Spinup Spinup Spindown Read Write

26

Microbenchmarks Microbenchmarks: Danger! : Danger!

• Need to evaluate over representative activity

100 200 300 400 500 600 700 800 900 10 20 30 40 50 60 Writeback Delay (s) File System Energy (J)

Default Flush on Write Flush on Spindown Flush on Both

27

Case Study: Network File System Case Study: Network File System

Proposed power management strategies:

• Flush on write

– When any data is sent to server, flush all data to server

• Flush on PSM

– Before network card transitions, flush all data to server

• Increase writeback delay

– Increase opportunities for bulk transfer

Ran 6 traces on BlueFS with Drive-Thru (8-28 hours)

28

Case Study: Network Storage Case Study: Network Storage

100 200 300 400 500 600 10 20 30 40 50 60 Writeback Delay (s) File System Energy (J)

Default Flush on Write Flush on PSM Flush on Both

• Complete evaluation over 13,000x faster than trace replay

29

User App

Improving Improving BlueFS BlueFS

Local Storage BlueFS BlueFS Server Network

30

User App

Improving Improving BlueFS BlueFS

Local Storage BlueFS BlueFS Server Network Flush on Spindown

31

User App

Improving Improving BlueFS BlueFS

Local Storage BlueFS BlueFS Server Network Flush on Spindown Reduced Writeback Delay From 30 to 2 seconds

32

Improving Improving BlueFS BlueFS

Modified BlueFS and ran 45-minute Purcell trace

• Implemented flush on spindown for disk

– File system energy reduced 12.4% – Interactive delay reduced 11.0%

• Reduced network writeback delay from 30 to 2 secs

– Safety improved for 1.8% file system energy cost

33

Related Work Related Work

• Disksim [Bucy 03] / Dempsey [Zedlewski 03]

– Detailed disk performance and power model

• QualNet [http://www.scalablenetworks.com]

NS2 [http://www.isi.edu/nsnam/ns]

– Detailed network performance model

• File-Cache-Content Detector [Arpaci-Dusseau 01]

34

Conclusion Conclusion

Drive-Thru is:

• Fast: Ran ext2 evaluation 40,000x faster
• Accurate: 3% error for disk file system energy
• Portable: < 1,000 lines of code

Case Study Insights:

• Increasing writeback delay a meager improvement
• Avoid writeback delay = disk spindown delay
• Flush on spindown effective for disk