Energy-aware Service Allocation for Cloud Computing Tobias Widmer 1 - - PowerPoint PPT Presentation

University of Hohenheim Information Systems 2

Energy-aware Service Allocation for Cloud Computing

Tobias Widmer1, Marc Premm1, Paul Karänke1,2

1 Universität Hohenheim 2 FZI Forschungszentrum Informatik, Karlsruhe

2

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Outline

1. Research Approach 2. Related Work 3. Model 4. Allocation Mechanism 5. Evaluation 6. Conclusion

3

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing
• 1. Research Approach
• Object of research
• Service network of customers and

cloud service providers (SPs), i.e., multiple SPs offer (standardised) cloud services to multiple customers

• Problem
• Energy-aware optimal service allocation
• Perspective
• Multiagent system technology
• Contribution:

Auction-based allocation mechanism as a heuristic (method artefact)

sp3 sp1 sp2 c1 c2

customers service providers

4

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

c1 c2

Basic Assumptions

• SPs offer (standardised, i.e., equal functional properties) cloud services of

specified energy consumption rate (metric: performance-per-watt)

[Sharma et al. 2006]

• SPs know their energy efficiency and capacity
• Energy-efficient cloud services are scarce
• Customers outsource infrastructure, platforms, and software applications

to cloud SPs such that physical machines can be switched off ( leading to higher energy efficiency)

• Customers and SPs model preferences via utility functions
• Service cost and energy efficiency correlate positively

sp1

customers service provider

5

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing
• 2. Related Work
• Energy-efficient cloud computing
• Most approaches consider energy efficiency within a single data

centre but do not take a global (social welfare) perspective

[Berl et al. 2009; Baliga et al. 2010; Bodenstein et al. 2011]

• Global cloud computing architecture on energy efficiency and its

impact on carbon dioxide emissions but requires central coordination entity [Garg et al. 2011]

• Agent-based cloud service allocation
• Central coordination entity required [Parkes & Shneidman 2004]
• Distributed coordination approaches do not consider energy

efficiency in the allocation rationale [Bo & Lesser, 2010; Bo et al. 2010]

6

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing
• 3. Model

Ej Energy efficiency of SP in performance/watt Wj Capacity of SP in performance cj(sk) Cost of SP to provide service of type sk Vi(hl

i)

Valuation of customer for host hl

i

w(sk) Computing requirement for service of type sk in performance

7

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Maximisation Problem

• Find allocation xijk that maximises the sum over all utilities, i.e.,

utilitarian social welfare [Chevaleyre et al. 2006] W* = max ∑ij (∑l (Vi(hl

i) xijk - ∑k cj (w(sk), ek )) xijk)

subject to for all j : ∑j ∑kw(sk) xijk ≤ Wj

• Computation problem is NP-complete [Garey & Johnson 1979]
• Heuristic is needed to approximate optimal solution

 We propose a distributed allocation mechanism

8

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing
• 4. Allocation Mechanism
• Combinatorial multi-attribute procurement auction

[Parkes & Kalagnanam 2005; Che 1993]

• Considers bundles of cloud services
• Considers price and energy efficiency as attributes
• Request bundle Ri = ((r1, ..., rn ), (r1, ..., rn ), ... , (r1, ..., rn ))

with rk = 1 iff service sk belongs to requested bundle, 0 otherw.

• The mechanism defines a score considering energy efficiency

and price:

σi(sk, ek, pk) = αi(w(sk) / ek) – βipk with scalars αi, βi

• Bid bundle Bj = ((σi(s1, e1, p1), e1), (σi(s2, e2, p2), e2), ... , (σi(sn,

en, pn), en)). If j does not bid for sk: (σi(s1, e1, p1), e1) := (0, 0)

9

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Second-score Auction

• Second-score auction: winning SP must match second highest

score [Che 1993]

• Example for a service of type s:
• A bids (5, 0.2) with score σi(s, 5, 0.2) = 10
• B bids (6, 0.15) with σi(s, 6, 0.15) = 9

 A is winner and is free to choose any tuple (e’, p’) such that σi(s, e’, p’) = 9

10

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Auction Protocol

customer service provider

Announce request for service types Inform winner and transfer second score Submit binding bids for service types Calculate optimal tuple (energy, price) based

• n second score: final contract is established

The proposed protocol is incentive-compatible for SP agents.

11

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing
• 5. Evaluation
• Objective: To demonstrate the efficiency of the artefact

(nearness to optimum)

• Evaluation method: Experimental evaluation – simulation with

artificial data (inspired by real-world data)

• Experimental setup
• From Amazon and SPECpower: 1 Mio. ssj_ops cost 6.75 € with

power consumption rate 186 Watt

• Poisson dist.: 10 hosts with 2 services/host
• 10 services: performance ranging from 0.3 to 2.1 Mio. ssj_ops
• 2..100 customer agents, 5 SP agents, 100 experiments
• Supply/demand balance at approx. 50 customer agents
• Different weights on energy and price (see score definition)

12

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Results (1): Energy-aware customers

Excess supply: customers easily find energy-efficient services but SP costs are high Services are scarce: high competition; customers are willing to forgo their energy-price preferences

13

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Results (2): Price-aware customers

Low competition: cheapest SPs are selected; results in higher social welfare Services are scarce: high competition; customers are willing to forgo their energy-price preferences

14

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Conclusion

• Findings
• A formal, game-theoretic framework for modeling energy-aware

cloud service allocation

• A novel distributed allocation mechanism that integrates energy

efficiency into the allocation rationale

• Limitations
• Considers two non-functional service properties only (energy

efficiency and price)

• Only 5 SP agents are considered in the experiment

15

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

Thank you for your attention!

Tobias Widmer Information Systems 2 University of Hohenheim tobias.widmer@uni-hohenheim.de

Förderkennzeichen 01ME11052 www.migrate-it2green.de

16

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing

References

• Baliga, B.J., Ayre, R.W.A., Hinton, K., Tucker, R.S. (2010). Green cloud computing: Balancing energy in

processing, storage and transport. Proceedings of the IEEE 99, 149–167.

• Berl, A., Gelenbe, E., Di Girolamo, M., Giuliani, G., De Meer, H., Dang, M.Q., Pentikousis, K. (2009). Energy-

efficient cloud computing. Comput. J. 53, 1045–1051.

• Bo, A., Lesser, V. (2010). Characterizing contract-based multi-agent resource allocation in networks. IEEE Sys.
• Man. Cybern. 40, 575–586.
• Bo, A., Lesser, V., Irwin, D., Zink, M. (2010). Automated negotiation with decommitment for dynamic resource

allocation in cloud computing. In: Ninth International Joint Conference on Autonomous Agents and Multi-Agent Systems (AAMAS’10), pp. 981–988.

• Bodenstein, C., Schryen, G., Neumann, D. (2012). Energy-aware workload management models for operation

cost reduction in data centers. European Journal of Operations Research 222, 157-167.

• Che, Y.K. (1993). Design competition through multidimensional auctions. RAND Journal of Economics, 24 (4),

668-680.

• Chevaleyre, Y., Dunne, P.E., Endriss, U., Lang, J., Lemaitre, M., Maudet, N., Padget, J., Phelps, S., Rodriguez-

Aguilar, J.A., Sousa, P. (2006). Issues in multiagent resource allocation. Informatica 30, 3–31.

• Fischer, K., Müller, J.P., Heimig, I. and Scheer, A. (1996). Intelligent agents in virtual enterprises. In

Proceedings 1st International Conference on the Practical Application of Intelligent Agents and Multi Agents Technology, London.

• Fuehrer, E. C. and Ashkanasy, N. M. (1998). The Virtual organization: defining a Weberian ideal type from the

inter-organizational perspective. Paper presented at the Annual Meeting of the Academy of Management, San Diego, CA.

17

• T. Widmer et al.: Energy-aware Service Allocation for Cloud Computing
• Garey, M.R., Johnson, D.S. (1979). Computers and Intractability: A Guide to the Theory of NP-Completeness. W.
• H. Freeman, New York
• Garg, S.K., Yeo, C.S., Buyya, R. (2011). Green cloud framework for improving carbon efficiency of clouds. In:

Jeannot, E., Namyst, R., Roman, J. (eds.) Euro-Par 2011. LNCS, vol. Part I, pp. 491–502, Springer, Heidelberg.

• Norman, T. J., Preece, A. D., Chalmers, S., Jennings, N. R., Luck, M., Dang, V. D., Nguyen, T. D., V. Deora, J.

S., Gray, W. A., and Fiddian, N. J. (2004). Agent-based formation of virtual organisations. International Journal

• f Knowledge Based Systems, 17 (2-4), 103-111.
• Parkes, D., Kalagnanam, J. (2005). Models for Iterative Multiattribute Procurement Auctions. Management

Science 51:435-451

• Parkes, D., Shneidman, J. (2004). Distributed implementations of Vickrey-Clarke-Groves mechanisms. In:

Proceedings of the 3rd International Conference on Autonomous Agents and Multiagent Systems (AAMAS’04),

• pp. 261–268, ACM, New York.
• Petersen, S. A. (2007). Virtual enterprise formation and partner selection: an analysis using case studies.

International Journal of Networking and Virtual Organisations, 4 (2), 201-215.

• Sharma, S., Hsu, C., Feng, W. (2006). Making a case for a green500 list. In: Proceedings of the Workshop on

High-Performance, Power-Aware Computing