Power-Saving in Large-Scale Storage Systems with Data Migration - - PowerPoint PPT Presentation

▶
power saving in large scale storage systems with data
SMART_READER_LITE
LIVE PREVIEW

Power-Saving in Large-Scale Storage Systems with Data Migration - - PowerPoint PPT Presentation

Dec 26, 2022 313 likes •794 views

Power-Saving in Large-Scale Storage Systems with Data Migration Koji Hasebe, Tatsuya Niwa, Akiyoshi Sugiki, and Kazuhiko Kato University of Tsukuba, Japan Background Power-saving in storage systems is a central issue. IT systems consume

slide-1
SLIDE 1

Power-Saving in Large-Scale Storage Systems with Data Migration

Koji Hasebe, Tatsuya Niwa, Akiyoshi Sugiki, and Kazuhiko Kato

University of Tsukuba, Japan

slide-2
SLIDE 2

Background

 IT systems consume 1-2% of the total energy in the world.

Green IT: A New Industry Shock Wave, Gartner Symp/ITxpo, 2007

 In large data centers, storage systems consume <40% of the

total power. StorageIO, Greg Sculz

 Power-saving in storage systems is a central issue.

slide-3
SLIDE 3

Previous Studies

Workload Low-power mode Peak time Off-peak time  In the literature…

 MAID [Colarelli-Grunwald, '02], PDC [Pinheiro-Bianchini, '04]  DIV [Pinheiro et al., '06], Pergamum [Storer et al., '08]  RIMAC [Yao-Wang, '06], eRAID [Wang-Zhu-Li, '08]  Hibernator [Zhu et al., '05], PARAID [Waddle et al. '07], etc.

 Commonly-observed technique:

slide-4
SLIDE 4

Previous Studies

Limitations:

  • Central controller to manage data accesses
  • Relatively small number of disks (up to several dozen)

Harnik et al. [IPDPS'09]

 Propose the efficient allocation of replicated data d1 d2 d3

slide-5
SLIDE 5

Previous Studies

Limitations:

  • Central controller to manage data accesses
  • Relatively small number of disks (up to several dozen)

Harnik et al. [IPDPS'09]

 Propose the efficient allocation of replicated data d1 d2 d3

slide-6
SLIDE 6

Previous Studies

Limitations:

  • Central controller to manage data accesses
  • Relatively small number of disks (up to several dozen)

Harnik et al. [IPDPS'09]

 Propose the efficient allocation of replicated data d1 d2 d3 Low-power mode

slide-7
SLIDE 7

Motivation and Objective

 Apply the skewing technique to large storage systems  Explore an efficient technique based on the data migration,

instead of the replication approach

slide-8
SLIDE 8

Motivation and Objective

 Apply the skewing technique to large storage systems  Explore an efficient technique based on the data migration,

instead of the replication approach

data data data data data

slide-9
SLIDE 9

Motivation and Objective

 Apply the skewing technique to large storage systems  Explore an efficient technique based on the data migration,

instead of the replication approach

data data data data data Low-power mode

slide-10
SLIDE 10

Central Idea (1)

Underlying System

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P11 P12 Block 1 Block 2 Block 3 Block 4 Parent Children Parent Child

 Assume that

 3 physical nodes are required at off-peak time  May increase up to four-fold

slide-11
SLIDE 11

Central Idea (1)

Underlying System

P1 P2 P3 V1 V2 V3 V4 V5 V6 V7 V8 V9 P4 P5 P6 P7 P8 P9 P10 P11 P12 Block 1 Block 2 Block 3 Block 4 V1 V2 V3 V4 V5 V6 V7 V8 V9

Managed by distributed hash table (DHT)

  • Cf. Chord [Stoica et al. '01]
slide-12
SLIDE 12

Central Idea (1)

Underlying System

V1 P1 V2 V3 V4 P2 V5 V6 V7 P3 V8 V9 P4 P5 P6 P7 P8 P9 P10 P11 P12 1 4 7 2 5 8 3 6 9 1 2 3 4 5 6 7 8 9 1 4 7 2 5 8 3 6 9 Block 1 Block 2 Block 3 Block 4

slide-13
SLIDE 13

Central Idea (2)

Migration of Virtual Nodes

V1 P1 V2 V3 V4 P2 V5 V6 V7 P3 V8 V9 P4 P5 P6 1 4 7 2 5 8 3 6 9 Block 1 Block 2

Overloaded

slide-14
SLIDE 14

Central Idea (2)

Migration of Virtual Nodes

V12/2 P1 V2 V3 V4 P2 V5 V6 V7 P3 V8 V9 P4 P5 P6 1 4 7 2 5 8 3 6 9 Block 1 Block 2

Overloaded

V11/2

Divide V1 into two

V9 V12/2 V11/2

slide-15
SLIDE 15

Central Idea (2)

Migration of Virtual Nodes

V12/2 P1 V2 V3 V4 P2 V5 V6 V7 P3 V8 V9 P4 P5 P6 4 7 2 5 8 3 6 9 Block 1 Block 2

Overloaded

V11/2

slide-16
SLIDE 16

Central Idea (2)

Migration of Virtual Nodes

V12/2 P1 V2 V3 V4 P2 V5 V6 V7 P3 V8 V9 P4 P5 P6 4 7 2 5 8 3 6 9 Block 1 Block 2

Overloaded

V11/2

slide-17
SLIDE 17

Central Idea (2)

Migration of Virtual Nodes

V12/2 P1 V2 V3 V42/2 P2 V5 V6 V7 P3 V8 V9 P4 P5 P6 4 7 2 5 8 3 6 9 Block 1 Block 2

Overloaded

V41/2 V11/2

slide-18
SLIDE 18

Central Idea (2)

Migration of Virtual Nodes

V12/2 P1 V2 V3 V42/2 P2 V5 V6 P3 V8 V9 P4 P5 P6 7 2 5 8 3 6 9 Block 1 Block 2

Overloaded

V41/2 V11/2 V72/2 V71/2

slide-19
SLIDE 19

Central Idea (2)

Migration of Virtual Nodes

V12/2 P1 V42/2 P2 P3 P4 P5 P6 Block 1 Block 2 V41/2 V11/2 V72/2 V22/2 V32/2 V52/2 V62/2 V82/2 V92/2 V71/2 V51/2 V21/2 V81/2 V61/2 V31/2 V91/2

slide-20
SLIDE 20

Central Idea (2)

Migration of Virtual Nodes

P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P11 P12 Block 1 Block 2 Block 3 Block 4 Parent Children Parent Child d d d d d d d d d d d d

slide-21
SLIDE 21

Power-Saving Algorithms

 Short-term optimization

 Extension  Reduction

 Long-term optimization

slide-22
SLIDE 22

Power-Saving Algorithm 1

Short-term Optimization (Extension)

 Procedure

  • 1. Each physical node checks its own workload.
  • 2. If the workload exceeds its capacity, then one of the

virtual nodes is split and migrated to its child block.

V12/2 V2 V3 V42/2 V5 V6 V72/2 V8 V9 V11/2 V41/2 (5) (8) V71/2 (2) (6) (9) (3) P1 P2 P3 P4 P5 P6 Parent Child

slide-23
SLIDE 23

Power-Saving Algorithm 1

Short-term Optimization (Extension)

 Notes:

 Reusing the stored data in the previous day enables the

migration by copying the difference.

 The mapping of virtual nodes effectively skews the workload. V12/2 V2 V3 V42/2 V5 V6 V72/2 V8 V9 V11/2 V41/2 (5) (8) V71/2 (2) (6) (9) (3) P1 P2 P3 P4 P5 P6 Parent Child

slide-24
SLIDE 24

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Problem

V1 V42/2 V5 V72/2 V9 V41/2 V71/2 P1 P2 P3 P4 P5 P6 Parent Child V22/2 V32/2 V62/2 V82/2 V21/2 V31/2 V62/2 V82/2

The remaining capacity of physical nodes The workload of each virtual node = 1 (1) (1) (2)

slide-25
SLIDE 25

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Wrong migration

V1 V42/2 V5 V72/2 V9 V41/2 V71/2 P1 P2 P3 P4 P5 P6 Parent Child V22/2 V32/2 V62/2 V82/2 V21/2 V31/2 V62/2 V82/2

(1) (1) (2)

slide-26
SLIDE 26

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Wrong migration

V1 V42/2 V5 V72/2 V9 V41/2 V71/2 P1 P2 P3 P4 P5 P6 Parent Child V22/2 V3 V6 V82/2 V21/2 V82/2

(1) (1) (2) (0) (0)

slide-27
SLIDE 27

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 The solution

V1 V42/2 V5 V72/2 V9 V41/2 V71/2 P1 P2 P3 P4 P5 P6 Parent Child V22/2 V32/2 V62/2 V82/2 V21/2 V31/2 V62/2 V82/2

(1) (1) (2) (0) (0) (0)

slide-28
SLIDE 28

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 The solution

V1 V4 V5 V7 V9 P1 P2 P3 P4 P5 P6 Parent Child V2 V32/2 V62/2 V8 V31/2 V62/2

(1) (1) (2) (0) (0) (0)

slide-29
SLIDE 29

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Procedure

  • 1. C → P: the information about the workloads for every virtual node
  • 2. P lists all possible combinations of a subset of physical nodes s.t. P can absorb

their virtual nodes

slide-30
SLIDE 30

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Procedure

  • 1. C → P: the information about the workloads for every virtual node
  • 2. P lists all possible combinations of a subset of physical nodes s.t. P can absorb

their virtual nodes P1 {P4, P5} P2 {P4, P5}, {P5, P6} P3 {P4, P5} Candidates

V1 V4 V5 V7 V9

P1 P2 P3

V2 V32/2 V6 V8

(1) (∞) (∞)

V32/2

P4 P5 P6

slide-31
SLIDE 31

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Procedure

  • 1. C → P: the information about the workloads for every virtual node
  • 2. P lists all possible combinations of a subset of physical nodes s.t. P can absorb

their virtual nodes P1 {P4, P5}, {P4, P6} P2 {P4, P5}, {P5, P6} P3 {P4, P5} Candidates

V1 V4 V5 V7 V9

P1 P2 P3

V22/2 V3 V6 V8

(1) (∞) (∞)

V22/2

P4 P5 P6

slide-32
SLIDE 32

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Procedure

  • 1. C → P: the information about the workloads for every virtual node
  • 2. P lists all possible combinations of a subset of physical nodes s.t. P can absorb

their virtual nodes

  • 3. P → C: the result of Step 2
  • 4. C calculates the intersection for all possible combinations of the results.

P1 {P4, P5}, {P4, P6} P2 {P4, P5}, {P5, P6} P3 {P4, P5} Candidates {P4, P5}

V1 V4 V5 V7 V9

P1 P2 P3

V2 V32/2 V62/2 V8

(1) (1) (2)

V32/2

P4 P5 P6

V62/2

slide-33
SLIDE 33

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Procedure

  • 1. C → P: the information about the workloads for every virtual node
  • 2. P lists all possible combinations of a subset of physical nodes s.t. P can absorb

their virtual nodes

  • 3. P → C: the result of Step 2
  • 4. C calculates the intersection for all possible combinations of the results.

P1 {P4, P5}, {P4, P6} P2 {P4, P5}, {P5, P6} P3 {P4, P5} Candidates {P4, P5}, {P5}

V1 V42/2 V5 V7 V9

P1 P2 P3

V2 V32/2 V6 V8

(1) (1) (2)

V32/2

P4 P5 P6

V41/2

slide-34
SLIDE 34

Power-Saving Algorithm 2

Short-term Optimization (Reduction)

 Procedure

  • 1. C → P: the information about the workloads for every virtual node
  • 2. P lists all possible combinations of a subset of physical nodes s.t. P can absorb

their virtual nodes

  • 3. P → C: the result of Step 2
  • 4. C calculates the intersection for all possible combinations of the results.

P1 {P4, P5}, {P4, P6} P2 {P4, P5}, {P5, P6} P3 {P4, P5} Candidates {P4, P5}, {P5}, {P4} Solution

V1 V4 V5 V7 V9

P1 P2 P3

V2 V32/2 V62/2 V8

(1) (1) (2)

V32/2

P4 P5 P6

V62/2

slide-35
SLIDE 35

Power-Saving Algorithm 3

Long-term Optimization

 To maintain effective power-saving, it requires load-

balancing in each block.

 Example:

V1 V2 V3 V4 V5 V6 V7 V8 V9 (1) (4) (5) (8) (7) (2) (6) (9) (3) P1 P2 P3 P4 P5 P6 Parent Child

High workload

slide-36
SLIDE 36

Power-Saving Algorithm 3

Long-term Optimization

 To maintain effective power-saving, it requires load-

balancing in each block.

 Example:

V4 V5 V6 V7 V8 V9 (4) (5) (8) (7) (6) (9) P1 P2 P3 P4 P5 P6 Parent Child V12/2 V22/2 V32/2 V11/2 V21/2 V31/2

Low workload

slide-37
SLIDE 37

Power-Saving Algorithm 3

Long-term Optimization

 To maintain effective power-saving, it requires load-

balancing in each block.

 Example:

V1 V2 V3 V4 V5 V6 V7 V8 V9 (1) (4) (5) (8) (7) (2) (6) (9) (3) P1 P2 P3 P4 P5 P6 Parent Child

slide-38
SLIDE 38

Power-Saving Algorithm 3

Long-term Optimization

 To maintain effective power-saving, it requires load-

balancing in each block.

 Example:

V1 V5 V9 V4 V2 V6 V7 V8 V3 (1) (4) (5) (8) (7) (2) (6) (9) (3) P1 P2 P3 P4 P5 P6 Parent Child

Load is balanced

slide-39
SLIDE 39

 Purposes

 Evaluate the efficiency of skewing the workload.  Evaluate the validity of long-term optimization.

 Simulation environment

39

Evaluation (Simulation)

Number of physical nodes 800 Number of virtual nodes 10,000 Term of simulation 1 day Migration condition Split：more than 90％ Merge：less than 70％ Workload of all virtual nodes Initially at its lowest，increased until middle of the day．Gap was sixfold. Virtual node groups Gap of the loads is twice.

slide-40
SLIDE 40

40

Simulation Results (Average load)

  • In the case

Without long-term optimization: 57-69% With Long-term optimization: 67-74%

Long-term optimization algorithm improves the average load as expected. Physical nodes run effectively, coping with the daily variation of workload.

Results

Time (hour) Average load of active physical nodes (%)

slide-41
SLIDE 41

Simulation Results (Active nodes)

Long-term optimization saves on Average: 7-14% Up to: 17-39%

41

Optimization improves the power consumption consistently and continually.

Results

Time (hour) The number of active physical nodes

slide-42
SLIDE 42

Purposes

 Verify the efficiency of load intensive at real machine.  Verify whether response time becomes below the desired time.

 Response time：from sending a request until the data were loaded into

memory in the server.

Experiment environment

42

Evaluation (Prototype implementation)

Number of physical nodes 40

  • Xeon 3.60GHZ CPU x2
  • Memory about 2GB
  • HDD(SCSI) 36GB

Number of Files 60,000 x 1MB (total 60GB) Term of experiment 1 day Migration condition Split：over 90％，Merge：under 70％ Workload of all virtual nodes Initially at its lowest，increased until middle

  • f the day．Gap is sixfold.

Virtual node groups Twice between two groups Amount of each migration 10% of all the data

slide-43
SLIDE 43

43

Response Time

  • Average response time

80msec

  • Maximum response time

534msec

Our algorithms can keep almost below desired response time.

Results

Time (hour) Response time per request (ms)

slide-44
SLIDE 44

44

Average Load

  • Overall average load:

67% of the capacity

Can also skew the workload effectively as the simulation.

Results

Time (hour) Load of active physical nodes (%)

slide-45
SLIDE 45

Number of Active Physical Nodes

  • Migration is done on

Average: 0.14 virtual nodes Maximum: 20 virtual nodes

Our system adjusts the number of physical nodes to the variation of workloads and reduces power effectively

Time (hour) The number of active physical nodes The number of migrations

slide-46
SLIDE 46

Conclusions

 Power-saving method for large-scale distributed storage

systems.

 Short/Long-term optimization algorithms for reducing power

consumption.

 Performance evaluation

 Simulation results showed that our method kept the workload

  • n

 Average: 67–74%

 Prototype implementation results showed that

 Overall Average load was: 67%  It can maintain a preferred response time

slide-47
SLIDE 47

Future work

 Implement replication mechanism to improve reliability.  Improve the long-term optimization algorithm.