a strategy-proof pricing scheme for multiple resource type...

25
A Strategy-proof Pricing Scheme for Multiple Resource Type Allocations Marian Mihailescu and Yong Meng Teo Department of Computer Science National University of Singapore

Upload: others

Post on 06-Jan-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

A Strategy-proof Pricing Scheme for Multiple Resource Type

Allocations

Marian Mihailescu and Yong Meng Teo Department of Computer Science National University of Singapore

Page 2: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Overview

•  Introduction and Related Work

•  Our Approach

•  Proposed Mechanism

•  Example

•  Simulation Results

•  Conclusions and Future Work

2 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 3: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Introduction

•  Large scale resource sharing   Grid

  Peer-to-peer

  Cloud Computing

•  Fundamental problem: resource allocation

•  Difficulty: rational users   Maximize their own interest in sharing

  Affect the performance of the system

3 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 4: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Mechanism Design

Problem

•  Mechanism design problem

  Outcome specification

  Set of user valuations for a specific outcome

Solution

•  Mechanism

  Social choice function f determines the outcome

  User payment

4

•  Provides a framework to design protocols that give rational agents incentives to interact in particular ways, such that social welfare is “maximized” at equilibrium

M = ( f , p1,..., pn )

f (t1…tn ) =maxo uii∑

pi

vi(ti,o)

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 5: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

•  Computational Efficiency   Optimal allocation requires

NP-complete algorithm

Desired Properties

Economic

•  Multiple Resource Types   A buyer request contains more than one

resource type

•  Strategy-proof   Users gain higher welfare from participating

and have no incentives to declare false information

•  Budget Balance   Sum of all user payments is 0, and allocations

do not result in deficit or surplus

•  Economic Efficiency   Resources are allocated to the user that values

them the most; total welfare is maximized

Computational

5 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 6: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

•  Computational Efficiency   Optimal allocation requires

NP-complete algorithm

Myerson-Satterthwite

Impossibility Theorem:

no mechanism achieves strategy-proof, budget balance

and economic efficiency

at the same time

Desired Properties

Economic

•  Multiple Resource Types   A buyer request contains more than one

resource type

•  Strategy-proof   Users gain higher welfare from participating

and have no incentives to declare false information

•  Budget Balance   Sum of all user payments is 0, and allocations

do not result in deficit or surplus

•  Economic Efficiency   Resources are allocated to the user that values

them the most; total welfare is maximized

Computational

6 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 7: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Related Work

Property Proportional

Share Bargaining Auctions

Combinatorial Auctions

Economic

Multiple Resource Types ✔ ✔ ✕ ✔

Strategy-proof ✕ ✕ ✔ ✔

Budget Balance ✔ ✔ ✔ ✕

Pareto Efficiency ✕ ✕ ✕ ✔

Computational

Algorithm Complexity low low low high

Tycoon (2004) [8] REXEC (2000) [4] Nimrod/G (2002) [2]

Popcorn (1998) [14] Spawn (1992) [18]

Mirage (2005) [3] Bellagio (2004) [1]

7 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 8: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Our Approach

8

Pareto Efficiency

Computational Efficiency

Budget Balance

Strategy-proof

Multiple Resource Types

Trade-off

Trade-off

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 9: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Market-based Resource Allocation Problem

•  Buyers submit requests for multiple resource types

•  Buyer private information   Maximum price the buyer is willing to pay such that, for each

resource type, resources are allocated to satisfy its request

•  Sellers publish each resource type separately

•  Seller private information for each resource type   Underlying costs for the respective resource type, such as power

consumption, bandwidth costs, etc.

•  For a particular request, the goal is to allocate resources such that the underlying costs are minimized

9 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 10: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Winner Determination

•  Centralized market-maker   Manage requests and resources

  Determine winners and compute payments

•  Reverse Auction based Winner Determination   Select one request (buyer winner)

  For each resource type in the request o  Select resources with minimum cost (seller winner)

10 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 11: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Payment Functions

•  Seller payment function

•  Buyer payment function €

ps =0 s does not contribute resources to allocate the request

−cM |s=∞ + cM |s=0 s contributes with resources to allocate the request

pb = − pss∈S∑

11

cM |s=∞ minimum cost to allocate the request without the resources of seller scM |s=0 minimum cost to allocate the request when the resource cost of seller s is 0VCG payment function: strategy-proof, Pareto-efficient, NOT budget-balanced

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 12: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Payment Functions

•  Seller payment function

•  Buyer payment function €

ps =0 s does not contribute resources to allocate the request

−cM |s=∞ + cM |s=0 s contributes with resources to allocate the request

pb = − pss∈S∑

12

cM |s=∞ minimum cost to allocate the request without the resources of seller scM |s=0 minimum cost to allocate the request when the resource cost of seller s is 0

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 13: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Achieved Properties

•  Multiple Resource Type

•  Seller Payment Function:   Strategy-proof   Economic Efficiency

•  Buyer Payment Function:   Strategy-proof (FCFS buyer requests)   Budget Balance

•  Computational Efficiency

13 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 14: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Example

14

S1 CPU $1

S2 DISK $2 S3

DISK $1

S2 CPU $2

B1 CPU+DISK $5

B2 CPU+DISK $6

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 15: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Proposed Mechanism

15

Market Maker

Resources

Requests

CPU [S1] = $1 CPU [S2] = $2

DISK [S2] = $2 DISK [S3] = $1

CPU + DISK [B1] = $5 CPU + DISK [B2] = $6

Winner Determination

Buyers Sellers

CPU DISK

B1 ($5) S1 ($1) S2 ($2)

B2 ($6) S2 ($2) S3 ($1)

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 16: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Payment Computation

Proposed Mechanism

16

Market Maker

Resources

Requests

CPU [S1] = $1 CPU [S2] = $2

DISK [S2] = $2 DISK [S3] = $1

CPU + DISK [B1] = $5 CPU + DISK [B2] = $6

Winner Determination

Agent Payment

S1 2 + 1 = 3 0 + 1 = 1 -3 + 1 = -2

S3 1 + 2 = 3 1 + 0 = 1 -3 + 1 = -2

B1 - - 2 + 2 = 4

cM |s=! cM |s=0

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 17: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Optimal Allocation

17

Market Maker

Resources

Requests

CPU [S1] = $1 CPU [S2] = $2

DISK [S2] = $2 DISK [S3] = $1

CPU + DISK [B1] = $5 CPU + DISK [B2] = $6

Winner Determination

Total Welfare Exchange

w/o S1 6 – 2 – 1 = 3 B2 buys from S2, S3

w/o S2 6 – 1 – 1 = 4 B2 buys from S1, S3

w/o S3 6 – 1 – 2 = 3 B2 buys from S1, S2

w/o B1 6 – 1 – 1 = 4 B2 buys from S1, S3

w/o B2 5 – 1 – 1 = 3 B1 buys from S1, S3

maximum 6 – 1 – 1 = 4 B2 buys from S1, S3

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 18: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Payment Computation

Optimal Allocation

18

Market Maker

Resources

Requests

CPU [S1] = $1 CPU [S2] = $2

DISK [S2] = $2 DISK [S3] = $1

CPU + DISK [B1] = $5 CPU + DISK [B2] = $6

Winner Determination

Agent Payment

S1 -1 – (4 – 3) = -2

S3 -1 – (4 – 3) = -2

B2 6 – (4 – 3) = 5

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 19: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Implementation

•  Discrete event auctions simulator

•  jCase – open-source combinatorial auctions simulator

•  FreePastry-based implementation on PlanetLab

Impact of Untruthful Users

5.5

6

6.5

7

7.5

8

8.5

1000 2000 3000 4000 5000 6000 7000 8000

Nu

mb

er o

f S

ucc

essf

ul

Req

ues

ts (

log

)

Number of Requests

truthful10% untruthful, 10% price change10% untruthful, 20% price change30% untruthful, 10% price change30% untruthful, 20% price change

19 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 20: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Comparison with Traditional One-sided Auctions

2

3

4

5

6

7

8

9

10

24 48 72 96 120 144 168

Num

ber

of

Succ

essf

ul

Req

ues

ts (

log)

Simulation Time (hours)

traditional auctions, 1 rttraditional auctions, 4 rttraditional auctions, 8 rt

traditional auctions, 16 rt

proposed mechanism, 1 rtproposed mechanism, 4 rtproposed mechanism, 8 rtproposed mechanism, 16 rt

20

Price Diversity

(%)

Successful Buyer Requests (%)

Traditional Auctions

Proposed Increase

(%)

Under-Demand

10 66.4 78.9 26.5

20 66.4 79 27.1

40 66.3 79 26.6

Balanced Market

10 54.7 69.2 18.9

20 54.6 69.3 19.0

40 54.5 69.0 19.2

Over-Demand

10 33.5 39.1 16.7

20 33.8 39.1 15.6

40 33.6 39.1 16.3

Page 21: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Comparison with Combinatorial Auctions

Pricing Mechanism

Number of Users

Properties Performance

IC BB EE Runtime Succ. Buyer Requests (%)

Alloc. Seller Items (%)

Combinatorial Auctions (VCG)

20 40 80

✔ ✔ ✔

-1,402 -1,544 -1,557

2,470 6,321

14,384

9.6 min 2.5 hrs

67.4 hrs

44.5 52.5 54.2

44.8 57.2 64.0

Combinatorial Auctions

(Threshold)

20 40 80

✕ ✕ ✕

5 9 6

2,491 6,223

14,567

9.8 min 2.5 hrs

49.5 hrs

44.4 49.6 58.8

48.3 59.8 65.1

Proposed

20 40 80 100 200 500

✔ ✔ ✔

✔ ✔ ✔

0 0 0 0 0 0

1,871 5,483

11,561 14,369 28,564 65,948

1 sec 3 sec 5 sec 7 sec

20 sec 1.9 min

36.3 48.5 52.8 54.1 53.5 52.6

32.5 55.3 68.0 71.6 76.5 80.2

Page 22: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

•  Scalability   Centralized market-maker that processes requests sequentially

  Vertical – increase the number of resource types

  Horizontal – increase the number of users

•  Monopolistic Sellers [Pham, H.N. et.al., An Approach to Vickrey-based Resource Allocation in the Presence of Monopolistic Sellers, In Proc. 7th Australasian Symposium on Grid Computing and e-Research (AusGrid 2009), pp. 77-83, Wellington, New Zealand]

Limitations

22

S1 CPU $1

S3 DISK $1

B1 CPU+DISK $5

S2 CPU $2

S2 DISK $2

B2 CPU+DISK $6

38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 23: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Conclusions

•  Resource pricing and allocation scheme that:   Allocates multiple resource types

  Provide incentives for rational buyers and sellers

  Achieves budget balance

  Computational efficiency

•  Future Work   Distributed pricing scheme – improve horizontal and

vertical scalability

23 38th International Conference on Parallel Processing, 22-25 September 2009, Vienna, Austria

Page 24: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Questions ?

[email protected]

Page 25: A Strategy-proof Pricing Scheme for Multiple Resource Type ...teoym/pub/09/ICPP2010-teoym-slides.pdf · Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource

Thank you !

[email protected]

Y. M. Teo and M. Mihailescu, A Strategy-proof Pricing Scheme for Multiple Resource Type Allocations, in Proceedings of 38th International Conference on Parallel Processing, pp. 172-179, IEEE Computer Society Press, Vienna, Austria, September 22-25, 2009