An Untold Story of Redundant Clouds: Making Your Service Deployment - - PowerPoint PPT Presentation

an untold story of redundant clouds making your service
SMART_READER_LITE
LIVE PREVIEW

An Untold Story of Redundant Clouds: Making Your Service Deployment - - PowerPoint PPT Presentation

An Untold Story of Redundant Clouds: Making Your Service Deployment Truly Reliable Ennan Zhai 1 , Ruichuan Chen 2 , David Isaac Wolinsky 1 , Bryan Ford 1 1 Yale University & 2 Bell Labs Road-Map Motivations Goal & Insight iRec


slide-1
SLIDE 1

An Untold Story of Redundant Clouds: Making Your Service Deployment Truly Reliable

Ennan Zhai1, Ruichuan Chen2, David Isaac Wolinsky1, Bryan Ford1

1Yale University & 2Bell Labs

slide-2
SLIDE 2

Road-Map

  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps
slide-3
SLIDE 3
  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps

Road-Map

slide-4
SLIDE 4

  • Application providers:
  • enjoy the simplicity of using the clouds
  • have no idea about what happen in the clouds
  • rent multiple clouds for redundancy

Background

slide-5
SLIDE 5

  • Application providers:
  • enjoy the simplicity of using the clouds
  • have no idea about what happen in the clouds
  • rent multiple clouds for redundancy

Background

slide-6
SLIDE 6

  • Application providers:
  • enjoy the simplicity of using the clouds
  • have no idea about what happen in the clouds
  • rent multiple clouds for redundancy

Background

slide-7
SLIDE 7

EC2 availability zone EC2 availability zone EC2 availability zone

Netflix Application Service

IaaS App

Example 1: Netflix

slide-8
SLIDE 8

iCloud Application Service

IaaS App

Amazon EC2 Service Microsoft Azure Service

Example 2: iCloud

slide-9
SLIDE 9

Email App

Cloud Provider A Cloud Provider B

Problem

slide-10
SLIDE 10

Email App

Cloud Provider A Cloud Provider B

Problem

Third-party infrastructure components

slide-11
SLIDE 11

Email App

Cloud Provider A Cloud Provider B

ISP B ISP A ISP C

Problem

Third-party infrastructure components

slide-12
SLIDE 12

Email App

Cloud Provider A Cloud Provider B

ISP B ISP A ISP C Power Source

Problem

Third-party infrastructure components

slide-13
SLIDE 13

Email App

Cloud Provider A Cloud Provider B

ISP B ISP A ISP C Power Source

Problem

Third-party infrastructure components

slide-14
SLIDE 14

Email App

Cloud Provider A Cloud Provider B

ISP B ISP A ISP C Power Source

Problem

Become unavailable !

Third-party infrastructure components

slide-15
SLIDE 15

Problem

slide-16
SLIDE 16
  • Cloud providers allocate or tolerate failures via:
  • diagnosis systems, e.g., Sherlock.
  • fault-tolerant systems, e.g., F10, Skute.

Existing Efforts

slide-17
SLIDE 17
  • Cloud providers allocate or tolerate failures via:
  • diagnosis systems, e.g., Sherlock.
  • fault-tolerant systems, e.g., F10, Skute.
  • Solving the problem after the outage occurs
  • There is no any effort before the problem occur

Existing Efforts

slide-18
SLIDE 18
  • Cloud providers allocate or tolerate failures via:
  • diagnosis systems, e.g., Sherlock.
  • fault-tolerant systems, e.g., F10, Skute.
  • Solving the problem after the outage occurs
  • We want to prevent the problem before the
  • utage occurs

Existing Efforts

slide-19
SLIDE 19
  • Cloud providers allocate or tolerate failures via:
  • diagnosis systems, e.g., Sherlock.
  • fault-tolerant systems, e.g., F10, Skute.
  • Solving the problem after the outage occurs
  • We want to prevent the problem before the
  • utage occurs

Existing Efforts

  • Recommending truly independent redundancy

services when deploying applications

slide-20
SLIDE 20
  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps

Road-Map

slide-21
SLIDE 21

Road-Map

  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps
slide-22
SLIDE 22

App Provider

Goal & Insight

Cloud A Cloud B Cloud C

slide-23
SLIDE 23

App Provider

Goal & Insight

Select two clouds for redundancy

Cloud A Cloud B Cloud C

slide-24
SLIDE 24

App Provider

Goal & Insight

Cloud A Cloud B Cloud C

A and B ?

slide-25
SLIDE 25

App Provider

Goal & Insight

Cloud A Cloud B Cloud C

B and C ?

slide-26
SLIDE 26

App Provider

Goal & Insight

Cloud A Cloud C

A and C ?

Cloud B

slide-27
SLIDE 27

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

Select two clouds for redundancy: A&B? B&C?

  • r A&C?
slide-28
SLIDE 28

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

slide-29
SLIDE 29

Cloud A Cloud B Cloud C App Provider Recommender

Goal & Insight

Assessing independence by the # of

  • verlapping components between clouds
slide-30
SLIDE 30

Cloud A Cloud B Cloud C App Provider Recommender

Goal & Insight

slide-31
SLIDE 31

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power B Power A

slide-32
SLIDE 32

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power B Power A ISP A Power A Power B

slide-33
SLIDE 33

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power B Power A ISP A Power A Power B

slide-34
SLIDE 34

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP A Power B ISP B Power A

slide-35
SLIDE 35

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP A Power B ISP B Power A ISP B Power A Power B

slide-36
SLIDE 36

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP A Power B ISP B Power A ISP B Power A Power B

slide-37
SLIDE 37

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A

slide-38
SLIDE 38

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

slide-39
SLIDE 39

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-40
SLIDE 40

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-41
SLIDE 41

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

=2

slide-42
SLIDE 42

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

=2

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-43
SLIDE 43

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-44
SLIDE 44

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

=1

slide-45
SLIDE 45

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP B Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

=1

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-46
SLIDE 46

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-47
SLIDE 47

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

=0

slide-48
SLIDE 48

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

ISP A Power A Power B ISP A Power B ISP B Power C Power A ISP B Power C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

=0

slide-49
SLIDE 49

App Provider Recommender

Goal & Insight

Cloud A Cloud B Cloud C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-50
SLIDE 50

Cloud A Cloud B Cloud C App Provider Recommender

  • 1. Cloud A, C 0
  • 2. Cloud B, C 1
  • 3. Cloud A, B 2

| |

Goal & Insight

Deployment

Ranking List

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

slide-51
SLIDE 51

Road-Map

  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps
slide-52
SLIDE 52

Road-Map

  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps
slide-53
SLIDE 53

Cloud Provider1 Cloud Provider2 Cloud Provider3 Recommender App Provider

Strawman Solution 1

slide-54
SLIDE 54

Cloud Provider1 Cloud Provider2 Cloud Provider3 Recommender App Provider

Privacy Concern!

Strawman Solution 1

slide-55
SLIDE 55

Cloud Provider1 Cloud Provider2 Cloud Provider3 Trusted Third Party App Provider

Strawman Solution 2

slide-56
SLIDE 56

Cloud Provider1 Cloud Provider2 Cloud Provider3 Trusted Third Party App Provider

It is hard to find!

Strawman Solution 2

slide-57
SLIDE 57

Cloud Provider1 Cloud Provider2 Cloud Provider3

Secure Multiparty Computation

App Provider

Strawman Solution 3

[Xiao et al, CCSW’13]

slide-58
SLIDE 58

Cloud Provider1 Cloud Provider2 Cloud Provider3 SMPC App Provider

SMPC is difficult to scale!

Strawman Solution 3

[Xiao et al, CCSW’13]

slide-59
SLIDE 59
  • The first cloud independence recommender sys:
  • achieving our goal
  • preserving privacy of each cloud provider
  • practical

Our Approach - iRec

slide-60
SLIDE 60

Preliminary background: PSI-CA

Our Approach - iRec

  • The first cloud independence recommender sys:
  • achieving our goal
  • preserving privacy of each cloud provider
  • practical
slide-61
SLIDE 61
  • Private set-intersection cardinality proposed by

[Freedman et al, EuroCrypt’04].

  • Allows k parties to compute the # of overlapping

elements without learning other information.

Preliminary: PSI-CA

slide-62
SLIDE 62

Preliminary: PSI-CA

  • Private set-intersection cardinality proposed by

[Freedman et al, EuroCrypt’04].

  • Allows k parties to compute the # of overlapping

elements without learning other information.

11 3 10 1 5 20 3 7 3

PSI-CA

slide-63
SLIDE 63

Preliminary: PSI-CA

11 3 10 1 5 20 3 7 3

PSI-CA

  • Private set-intersection cardinality proposed by

[Freedman et al, EuroCrypt’04].

  • Allows k parties to compute the # of overlapping

elements without learning other information.

slide-64
SLIDE 64

11 3 10 1 5 20 3 7 3

PSI-CA

One overlapping element

Preliminary: PSI-CA

One overlapping element One overlapping element

  • Private set-intersection cardinality proposed by

[Freedman et al, EuroCrypt’04].

  • Allows k parties to compute the # of overlapping

elements without learning other information.

slide-65
SLIDE 65

11 3 10 1 5 20 3 7 3

PSI-CA

But I do not know which element is overlapping

Preliminary: PSI-CA

But I do not know which element is overlapping But I do not know which element is overlapping

  • Private set-intersection cardinality proposed by

[Freedman et al, EuroCrypt’04].

  • Allows k parties to compute the # of overlapping

elements without learning other information.

slide-66
SLIDE 66

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

slide-67
SLIDE 67

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240

slide-68
SLIDE 68

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {E(1), E(-21), E(128), E(-240)}

slide-69
SLIDE 69

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {E(P(1)), E(P(4)), E(P(6)), E(P(2))} {E(1), E(-21), E(128), E(-240)}

slide-70
SLIDE 70

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {E(P(1)), E(P(4)), E(P(6)), E(P(2))} {E(1), E(-21), E(128), E(-240)}

slide-71
SLIDE 71

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {E(P(1)), E(P(4)), E(P(6)), E(P(2))} {E(1), E(-21), E(128), E(-240)}

slide-72
SLIDE 72

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {E(-132), E(0), E(-12), E(-60)} {E(1), E(-21), E(128), E(-240)}

slide-73
SLIDE 73

1 4 6 2

  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A Data Set B

Alice Bob

Preliminary: PSI-CA

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {E(-132), E(0), E(-12), E(-60)} {E(1), E(-21), E(128), E(-240)}

slide-74
SLIDE 74
  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A

Alice

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {-132, 0, -12, -60}

Preliminary: PSI-CA

slide-75
SLIDE 75
  • Alice and Bob has set A and B respectively and Alice wants to jointly

compute |A B|.

  • Alice makes a polynomial P whose roots are the elements of data set A.
  • Alice encrypts the coefficients of P and sends them to Bob. Note that Alice sends

homomorphic encryptions of the coefficients to Bob.

  • Bob evaluates P(Bi) for each element in data set B.
  • Bob returns the encrypted evaluations to Alice.
  • Alice decrypts it and counts the number of zeroes.

Data Set A

Alice

P = (X-12)(X-5)(X-4) = x3-21x2+128X-240 {-132, 0, -12, -60}

Preliminary: PSI-CA

Result is: 1

slide-76
SLIDE 76

11 3 10 1 5 20 3 7 3

PSI-CA

One overlapping element

Preliminary: PSI-CA

One overlapping element One overlapping element

slide-77
SLIDE 77

Cloud A Cloud B Cloud C App Provider iRec

Our Approach - iRec

ISP A Power B ISP B Power C Power A

slide-78
SLIDE 78

Cloud A Cloud B Cloud C App Provider iRec

Our Approach - iRec

ISP A Power B ISP B Power C Power A Select two clouds for redundancy: A&B? B&C?

  • r A&C?
slide-79
SLIDE 79

Cloud A Cloud B Cloud C App Provider iRec

Step 1

ISP A Power B ISP B Power C Power A

slide-80
SLIDE 80

Cloud A Cloud B Cloud C App Provider iRec

Step 2

ISP A Power B ISP B Power C Power A

slide-81
SLIDE 81

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Step 3

iRec

ISP A Power A Power B ISP B Power A Power B ISP B Power C

slide-82
SLIDE 82

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Step 3

iRec

PSI-CA PSI-CA

ISP A Power A Power B ISP B Power A Power B ISP B Power C

slide-83
SLIDE 83

Cloud A Cloud B Cloud C App Provider iRec

Step 4

ISP A Power B ISP B Power C Power A

slide-84
SLIDE 84

App Provider Recommender

Step 5

Cloud A Cloud B Cloud C

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A Power B ISP B Power C Power A

slide-85
SLIDE 85

Cloud A Cloud B Cloud C App Provider Recommender

  • 1. Cloud A, C 0
  • 2. Cloud B, C 1
  • 3. Cloud A, B 2

| |

Step 5

Deployment

Ranking List

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A Power B ISP B Power C Power A

slide-86
SLIDE 86
  • Different infrastructure components play different

roles in the clouds

  • Power source might be much more likely to fail

than ISPs

An Improvement Version

slide-87
SLIDE 87
  • Different infrastructure components play different

roles in the clouds

  • Power source might be much more likely to fail

than ISPs

An Improvement Version

slide-88
SLIDE 88

An Improvement Version

  • We propose an improvement version
  • Using Weighted PSI-CA (W-PSI-CA) to instead of

PSI-CA in Step2

  • No other improvement
  • Different infrastructure components play different

roles in the clouds

  • Power source might be much more likely to fail

than ISPs

slide-89
SLIDE 89

An Improvement Version

  • We propose an improvement version
  • Using Weighted PSI-CA (W-PSI-CA) to instead of

PSI-CA in Step3

  • No other improvement
  • Different infrastructure components play different

roles in the clouds

  • Power source might be much more likely to fail

than ISPs

slide-90
SLIDE 90

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

iRec

Recall: Step 3

PSI-CA PSI-CA

ISP A Power A Power B ISP B Power A Power B ISP B Power C

slide-91
SLIDE 91

Cloud A Cloud B

ISP A Power B ISP B Power A

PSI-CA

ISP A Power A Power B ISP B Power A Power B

Recall: Step 3

Result is 2

slide-92
SLIDE 92

Cloud A Cloud B

ISP A Power B ISP B Power A

Using W-PSI-CA

slide-93
SLIDE 93

Cloud A Cloud B

ISP A Power B ISP B Power A ISP A Power A Power B ISP B Power A Power B

Using W-PSI-CA

slide-94
SLIDE 94

Cloud A Cloud B

ISP A Power B ISP B Power A

Using W-PSI-CA

ISP A 1 Power A 2 Power B 2 ISP B 1 Power A 2 Power B 2

slide-95
SLIDE 95

Cloud A Cloud B

ISP A Power B ISP B Power A

Using W-PSI-CA

ISP A 1 Power A 2 Power B 2 ISP B 1 Power A 2 Power B 2

Weights

slide-96
SLIDE 96

Cloud A Cloud B

ISP A Power B ISP B Power A

Using W-PSI-CA

ISP A Power A Power A Power B Power B ISP B Power A Power A Power B Power B

ISP A 1 Power A 2 Power B 2 ISP B 1 Power A 2 Power B 2 DSI DSI

slide-97
SLIDE 97

Cloud A Cloud B

ISP A Power B ISP B Power A

Using W-PSI-CA

ISP A Power A Power A Power B Power B ISP B Power A Power A Power B Power B

ISP A 1 Power A 2 Power B 2 ISP B 1 Power A 2 Power B 2 DSI DSI

slide-98
SLIDE 98

Cloud A Cloud B

ISP A Power B ISP B Power A

Using W-PSI-CA

PSI-CA

ISP A Power A Power A Power B Power B ISP B Power A Power A Power B Power B

DSI DSI

slide-99
SLIDE 99

Cloud A Cloud B

ISP A Power B ISP B Power A

Using W-PSI-CA

PSI-CA

Result is 4

ISP A Power A Power A Power B Power B ISP B Power A Power A Power B Power B

DSI DSI

slide-100
SLIDE 100

Case Study

slide-101
SLIDE 101

Cloud A Cloud B Cloud C App Provider iRec

Step 1

ISP A Power B ISP B Power C Power A Select two clouds for redundancy: A&B? B&C?

  • r A&C?
slide-102
SLIDE 102

Cloud A Cloud B Cloud C App Provider iRec

Step 1

ISP A Power B ISP B Power C Power A

slide-103
SLIDE 103

Cloud A Cloud B Cloud C App Provider iRec

Step 2

ISP A Power B ISP B Power C Power A

Weights Weights Weights

slide-104
SLIDE 104

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A 3 Power A 1 Power B 1 ISP B 3 Power A 1 Power B 1 ISP B 3 Power C 1

Step 3 & 4 with W-PSI-CA

iRec

slide-105
SLIDE 105

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A 3 Power A 1 Power B 1 ISP B 3 Power A 1 Power B 1 ISP B 3 Power C 1

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power A Power B ISP B ISP B ISP B Power C

Step 3 & 4 with W-PSI-CA

iRec

slide-106
SLIDE 106

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power A Power B ISP B ISP B ISP B Power C

Step 3 & 4 with W-PSI-CA

iRec

slide-107
SLIDE 107

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power C

PSI-CA

Step 3 & 4 with W-PSI-CA

iRec

slide-108
SLIDE 108

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power C

PSI-CA

Step 3 & 4 with W-PSI-CA

iRec

slide-109
SLIDE 109

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 1 Cloud A, B 2

Deployment | |

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power A Power B ISP B ISP B ISP B Power C

PSI-CA

Step 3 & 4 with W-PSI-CA

iRec

slide-110
SLIDE 110

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 3 Cloud A, B 2

Deployment | |

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power A Power B ISP B ISP B ISP B Power C

PSI-CA

Step 3 & 4 with W-PSI-CA

iRec

slide-111
SLIDE 111

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 3 Cloud A, B 2

Deployment | |

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power A Power B ISP B ISP B ISP B Power C

PSI-CA

Step 3 & 4 with W-PSI-CA

iRec

slide-112
SLIDE 112

App Provider Cloud A Cloud B Cloud C

ISP A Power B ISP B Power C Power A

Cloud A, C 0 Cloud B, C 3 Cloud A, B 2

Deployment | |

ISP A ISP A ISP A Power A Power B ISP B ISP B ISP B Power A Power B ISP B ISP B ISP B Power C

PSI-CA

Step 3 & 4 with W-PSI-CA

iRec

slide-113
SLIDE 113

App Provider Cloud A Cloud B Cloud C

Cloud A, C 0 Cloud B, C 3 Cloud A, B 2

Deployment | |

Step 5

iRec

slide-114
SLIDE 114

App Provider Cloud A Cloud B Cloud C

Cloud A, C 0 Cloud B, C 3 Cloud A, B 2

Deployment | |

Step 5

iRec

slide-115
SLIDE 115

App Provider Cloud A Cloud B Cloud C

Cloud A, C 0 Cloud A, B 2 Cloud B, C 3

Deployment | |

Step 5

iRec

slide-116
SLIDE 116

Cloud A Cloud B Cloud C App Provider

  • 1. Cloud A, C 0
  • 2. Cloud A, B 2
  • 3. Cloud B, C 3

| |

Deployment

Ranking List

Cloud A, C 0 Cloud A, B 2 Cloud B, C 3

Deployment | |

Step 5

iRec

slide-117
SLIDE 117

App Provider

  • 1. Cloud A, C 0
  • 2. Cloud A, B 2
  • 3. Cloud B, C 3

| |

Deployment

Ranking List

Step 5

  • 1. Cloud A, C 0
  • 2. Cloud B, C 1
  • 3. Cloud A, B 2

| |

Deployment

V.S.

Ranking list with W-PSI-CA Ranking list with PSI-CA iRec

slide-118
SLIDE 118

Road-Map

  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps
slide-119
SLIDE 119

Road-Map

  • Motivations
  • Goal & Insight
  • iRec System
  • Next Steps
slide-120
SLIDE 120

Next Steps

  • Can we provide stronger privacy preservation?
  • Do cloud providers have incentives to join?
  • Will the clouds behave honestly?
  • Can we make iRec more scalable?
  • How do we evaluate iRec with realistic cloud

dependency datasets?

slide-121
SLIDE 121

Next Steps

  • Can we provide stronger privacy preservation?
  • Do cloud providers have incentives to join?
  • Will the clouds behave honestly?
  • Can we make iRec more scalable?
  • How do we evaluate iRec with realistic cloud

dependency datasets?

slide-122
SLIDE 122

Next Steps

  • Can we provide stronger privacy preservation?
  • Do cloud providers have incentives to join?
  • Will the clouds behave honestly?
  • Can we make iRec more scalable?
  • How do we evaluate iRec with realistic cloud

dependency datasets?

slide-123
SLIDE 123

Next Steps

  • Can we provide stronger privacy preservation?
  • Do cloud providers have incentives to join?
  • Will the clouds behave honestly?
  • Can we make iRec more scalable?
  • How do we evaluate iRec with realistic cloud

dependency datasets?

slide-124
SLIDE 124

Thanks! Questions?