for and against the Cloud Roger Wattenhofer ETH Zurich Distributed - - PowerPoint PPT Presentation

▶
for and against
SMART_READER_LITE
LIVE PREVIEW

for and against the Cloud Roger Wattenhofer ETH Zurich Distributed - - PowerPoint PPT Presentation

Mar 21, 2024 220 likes •1.7k views

Algorithms for and against the Cloud Roger Wattenhofer ETH Zurich Distributed Computing Group Disclaimer SenSys OSDI HotNets AAAI PODC Mobicom STOC FOCS SIGCOMM ICALP SPAA SODA EC Algorithms for the Cloud Algorithms for the

slide-1
SLIDE 1 ETH Zurich – Distributed Computing Group Roger Wattenhofer

Algorithms for and against the Cloud

slide-2
SLIDE 2

PODC SODA STOC FOCS ICALP SPAA EC SenSys OSDI Mobicom AAAI SIGCOMM HotNets

Disclaimer

slide-3
SLIDE 3

Algorithms for the Cloud

slide-4
SLIDE 4

Algorithms for the Cloud

Infrast frastru ructure cture

slide-5
SLIDE 5

Algorithms for the Cloud

just perfe fect ct

slide-6
SLIDE 6

Algorithms for the Cloud

Infrast frastru ructure cture

slide-7
SLIDE 7
slide-8
SLIDE 8
slide-9
SLIDE 9
slide-10
SLIDE 10
slide-11
SLIDE 11
slide-12
SLIDE 12

Find balanced separator

  • f minimum size 𝐿.
slide-13
SLIDE 13

Find balanced separator

  • f minimum size 𝐿.
slide-14
SLIDE 14

Find balanced separator

  • f minimum size 𝐿.
slide-15
SLIDE 15

Our result: almost linear time algorithm for small 𝐿.

[Brandt, W., 2017]

Find balanced separator

  • f minimum size 𝐿.
slide-16
SLIDE 16

[Brandt, W., 2017]

…in a boring way

Find balanced separator

  • f minimum size 𝐿.

Our result: almost linear time algorithm for small 𝐿.

slide-17
SLIDE 17

Algorithms for the Cloud

just perfe fect ct

slide-18
SLIDE 18

GPS for the Cloud

slide-19
SLIDE 19

Just record 1ms of raw data

slide-20
SLIDE 20

Coarse Time Navigation Exhaustive Search in Area

slide-21
SLIDE 21

Also Robust to GPS Spoofing

slide-22
SLIDE 22

Algorithms for the Cloud

just perfe fect ct

slide-23
SLIDE 23

$100B Revenue ¾ Online

slide-24
SLIDE 24

Match Players Fast Waiting is Boooooring Match Players Well Similar Rating, Location, etc.

Online Two Player Games

slide-25
SLIDE 25

Min-Cost Perfect Matching With Delays (MPMD)

slide-26
SLIDE 26

MPMD Example

time rating (space)

slide-27
SLIDE 27

MPMD Example

time rating (space)

slide-28
SLIDE 28

MPMD Example

time rating (space)

slide-29
SLIDE 29

MPMD Example

time rating (space)

slide-30
SLIDE 30

MPMD Example

time rating (space) time cost space cost

slide-31
SLIDE 31

MPMD Example

time rating (space) time cost space cost

slide-32
SLIDE 32

MPMD Example

time rating (space) time cost space cost

slide-33
SLIDE 33

MPMD Example

time rating (space) time cost space cost

slide-34
SLIDE 34

Haste Makes Waste!

slide-35
SLIDE 35

MPMD Example

time rating (space) time cost space cost

slide-36
SLIDE 36

MPMD Example

time rating (space)

slide-37
SLIDE 37

MPMD Example

time rating (space)

slide-38
SLIDE 38

MPMD Example

time rating (space) algorithm cost

  • ptimal cost
slide-39
SLIDE 39
slide-40
SLIDE 40

[Wang et al., 2018] …

slide-41
SLIDE 41

The 𝑃(log 𝑜) Algorithm

slide-42
SLIDE 42

Approximate Metric by Tree

Leaves = Nodes in Metric Space

[Fakcharoenphol, Rao, Talwar 2004], [Bansal, Buchbinder, Gupta, Naor 2015]

Height = 𝑃(log𝑜) E[Distortion] = 𝑃(log𝑜) 𝑥

slide-43
SLIDE 43

Algorithm

slide-44
SLIDE 44

Algorithm

slide-45
SLIDE 45

Algorithm

= 𝑥

slide-46
SLIDE 46

Algorithm

slide-47
SLIDE 47

Algorithm

slide-48
SLIDE 48

Algorithm

slide-49
SLIDE 49

Algorithm

slide-50
SLIDE 50

Algorithm

slide-51
SLIDE 51

Algorithm

slide-52
SLIDE 52

Algorithm

slide-53
SLIDE 53

Proof

slide-54
SLIDE 54

Proof

slide-55
SLIDE 55

Proof

slide-56
SLIDE 56

Proof

Total space cost = σ

slide-57
SLIDE 57

Proof

slide-58
SLIDE 58

Proof

slide-59
SLIDE 59

Proof

For each pair at least one timer running Total time cost ≤ 2 σ

slide-60
SLIDE 60

Total Algorithm Cost = 𝑃(σ )

slide-61
SLIDE 61

What about OPT?

slide-62
SLIDE 62

Proof

slide-63
SLIDE 63

Proof

time ALG OPT

slide-64
SLIDE 64

Proof

time ALG OPT

slide-65
SLIDE 65

Proof

time ALG OPT

slide-66
SLIDE 66

Proof

time ALG OPT time ALG OPT

  • r
slide-67
SLIDE 67

Proof

time ALG OPT time ALG OPT

  • r

cost = cost

slide-68
SLIDE 68

Done?

slide-69
SLIDE 69

Just One Little Thing…

slide-70
SLIDE 70

Proof

time ALG OPT

slide-71
SLIDE 71

Proof

time ALG OPT

slide-72
SLIDE 72

Proof

time ALG OPT

slide-73
SLIDE 73

Proof

time ALG OPT

slide-74
SLIDE 74

Proof

time ALG OPT

slide-75
SLIDE 75

Proof

time ALG OPT

slide-76
SLIDE 76

Proof

time ALG OPT

slide-77
SLIDE 77

Proof

time ALG OPT

slide-78
SLIDE 78

Proof

time ALG OPT

slide-79
SLIDE 79

OPT has an easy time…

slide-80
SLIDE 80

… but only every other phase!

slide-81
SLIDE 81

Total OPT Cost = 𝛻(σ )

slide-82
SLIDE 82

Where is the log 𝑜 coming from?

Height = 𝑃(log𝑜) for time E[Distortion] = 𝑃(log 𝑜) for space

slide-83
SLIDE 83
slide-84
SLIDE 84

Algorithms against the Cloud

slide-85
SLIDE 85
slide-86
SLIDE 86
slide-87
SLIDE 87

2008

slide-88
SLIDE 88

Blockchain

slide-89
SLIDE 89
slide-90
SLIDE 90

Blockchain Basics

slide-91
SLIDE 91

Transaction

slide-92
SLIDE 92

Transaction

slide-93
SLIDE 93

Transaction

slide-94
SLIDE 94

Transaction

slide-95
SLIDE 95

Block

slide-96
SLIDE 96

Blockchain

slide-97
SLIDE 97

Blockchain is Replicated

slide-98
SLIDE 98

Distributed Systems & Cryptography (1982) (1976)

Blockchain

slide-99
SLIDE 99

Distributed Systems & Cryptography Fault-Tolerance & Digital Signatures

Blockchain

slide-100
SLIDE 100

Rule of Thumb Blockchains* may disrupt your business if you use signatures.

*or blockchain-like tech

slide-101
SLIDE 101

Blockchain Variants

slide-102
SLIDE 102
slide-103
SLIDE 103

Permissionless / Open

slide-104
SLIDE 104

Permissioned / Closed

slide-105
SLIDE 105

Multiple Participants? Writers Known? No Blockchain (use DB/Cloud) Permissionless Blockchain Permissioned Blockchain yes yes no no

slide-106
SLIDE 106

The Seven Blockchain Dimensions

Persistence Privacy Energy

slide-107
SLIDE 107

Persistence Database Provable Correct Byzantine Fault-Tolerance

Blockchain

Immutable Crash

slide-108
SLIDE 108

1 hour 1 minute Speed

Blockchain

10 tx/s 10m tx/s Throughput 1 second 10k tx/s

slide-109
SLIDE 109

10 nodes 100 nodes Scalability

Blockchain

1000 nodes

slide-110
SLIDE 110

Energy Consumption

slide-111
SLIDE 111
slide-112
SLIDE 112

Market / Energy Value ≈ 12 GW $1M/h $0.08/kWh

Economic Incentives

slide-113
SLIDE 113

Hashrate ∙ Energy/Hash ≈ 1.3 GW 13 ∙ 109 GH/s 0.1 J/GH

Proof of Work

slide-114
SLIDE 114

The Seven Blockchain Dimensions

Persistence Privacy Energy

slide-115
SLIDE 115

What About Privacy?

slide-116
SLIDE 116

It’s Complicated.

slide-117
SLIDE 117

Anonymity/Public Identity/Private

Privacy

slide-118
SLIDE 118

Research Issues

Persistence Privacy Energy

Solution to “many” problems: “Layer 2” Plus: crypto, language (smart contracts), game theory, measurements, …

slide-119
SLIDE 119

eMoney

slide-120
SLIDE 120

Permissioned Blockchain & Payment Network

slide-121
SLIDE 121

Permissioned Blockchain

slide-122
SLIDE 122

Payment Network

slide-123
SLIDE 123

Bitcoin

Anonymity Open/Anarchic Blockchain Eventual Consistency Proof-of-Work

eMoney

Accountability Closed/Private Paxos, PBFT, … Strong Consistency Central Banks

slide-124
SLIDE 124

eVoting

slide-125
SLIDE 125

What’s Wrong with Paper?

slide-126
SLIDE 126

Cost

slide-127
SLIDE 127

Verifiability

slide-128
SLIDE 128

Identity Swapper Identity Mixer …

Anonymity

slide-129
SLIDE 129

Election Help

slide-130
SLIDE 130

Democracy Beyond Yes or No

2025 Wie viel sollen die SRG-Gebühren pro Jahr kosten?

42.-

slide-131
SLIDE 131

Don’t bring a Blockchain to a Gunfight

slide-132
SLIDE 132

So what’s new, really?

slide-133
SLIDE 133 137

Classical Adversary

timing crashes

  • mission

Byzantine Hello World!

Now solve consensus

slide-134
SLIDE 134 138

meltdown spectre re-entrancy rowhammer Здравствуйте!

Now hold an election Modern Adversary

slide-135
SLIDE 135

Hype … and Criticism

“First practical solution to a longstanding problem in computer science, Byzantine Generals.” “Satoshi solved a problem that academic computer scientists thought was impossible” “Bitcoin is digital gold, it will put us back onto a sound monetary policy” “Bitcoin will end wars” “A non-deliberate Ponzi scheme” “It’s yet another eventually consistent database” “Flawed technology, inherently limited in scale and performance” “Unlikely to impact the finance sector”

slide-136
SLIDE 136

Cloud vs. Blockchain Would you rather fight…?

slide-137
SLIDE 137

Cloud vs. Blockchain Would you rather trust…?

miner relay developer user hodler big corporation

slide-138
SLIDE 138

Cloud vs. Blockchain Thanks to lots of hardware…

slide-139
SLIDE 139

“We at big corp will run your blockchain in our cloud!”

slide-140
SLIDE 140

What’s this?

slide-141
SLIDE 141

A Blockchain?

slide-142
SLIDE 142

A Cloud?

slide-143
SLIDE 143

A Distributed System!

slide-144
SLIDE 144

Summary

slide-145
SLIDE 145

Thank You!

Questions & Comments?

www.disco.ethz.ch

Thanks to my co-authors Vertex Separators: Sebastian Brandt Online With Delay: Yuval Emek, Shay Kutten Cloud GPS: Manuel Eichelberger

slide-146
SLIDE 146 Abstract:

Algorithms interact in two main ways with the cloud. There exist algorithms which are tailored for the cloud, for which the cloud is the perfect environment. Moreover, the cloud may also benefit from optimization algorithms, algorithms that make the cloud more efficient. The AlgoCloud program features papers which roughly fit one of the two, and I will also give a few examples in the first part of my talk. Apart from these algorithms for the cloud, I will also talk about algorithms against the cloud. Recently, blockchains are hyped to be a cloud competitor, sometimes even a cloud killer. In the second part of my talk we will discuss whether there is some truth to whether blockchains are going to threaten the successful cloud paradigm.