Towards Energy-efficient Cloud Computing Michael Maurer Distributed Systems Group TU Vienna, Austria maurer@infosys.tuwien.ac.at http://www.infosys.tuwien.ac.at/staff/maurer/
Distributed Systems Group Schahram Dustdar WWTF project: Foundations of Self-Governing ICT Infrastructures Ivona Brandic Vincent Emeakaroha Ivan Breskovic 2
HALEY Holistic Energy Efficient Approach for the Management of Hybrid Clouds Ivona Brandic Toni Mastelic Drazen Lucanin Wissenschaftspreis der TU Wien 2011 3
Collaborations Simulation theories for knowledge management: Manchester, R. Sakellariou Energy-efficient large scale distributed systems: Université Paul Sabatier, Toulouse Jean-Marc Pierson Concepts for performance predictions: UPC Barcelona, J. Torres Machine Learning Service virtualization: Sztaki, Budapest A. Kertész, G. Kecskeméti FOSII Scientific worklflow in the Cloud: BOKU Vienna, David Kreil, Pawel Labaj Hardware + real world apps for experimental testbeds: Porto Alegre, M. Netto, C. De Rose 4 Network economics: Seoul, J. Altmann Prototypes for app deployment: Melbourne, R. Calheiros, R. Buyya
Green Cloud Computing Goals Provide computing as a utility Dynamic Reliable Scalable Pay-as-you-go Energy efficient Best effort vs. keep to energy caps 5
Green Cloud Computing Methods, technologies Virtualization Monitoring Scheduling Experience on Clusters, Grids Electronic market theory Optimization theory 6
SaaS, PaaS, IaaS Software as a Service: Platform as a Service: Infrastructure as a Service: 7
FoSII infrastructure Autonomic Computing, Goals Goals: Few SLA violations High utilization of resources Energy-efficient usage Few reconfigurations actions (e.g., migrations) Good power management Service Level Agreement (SLA) CPU Power 512 MIPS Memory 1024 MB Storage 1000 GB Incoming Bandwidth 10 Mbit/s Outgoing Bandwidth 20 Mbit/s 8
Escalation levels Complexity of management 9
Complexity of management Escalation levels Do nothing 10
Complexity of management Escalation levels Do nothing 1. Change VM configuration 11
Complexity of management Escalation levels Do nothing 1. Change VM configuration 2. Migrate applications from one VM to another. 12
Complexity of management Escalation levels Do nothing 1. Change VM configuration 2. Migrate applications from one VM to another. 3. Migrate one VM from one PM to another or create new VM on appropriate PM. (BIP problem, NP-hard) 13
Complexity of management Escalation levels Do nothing 1. Change VM configuration 2. Migrate applications from one VM to another. 3. Migrate one VM from one PM to another or create new VM on appropriate PM. (BIP problem, NP-hard) 4. Turn on/off PM. 14
Complexity of management Escalation levels Do nothing 1. Change VM configuration 2. Migrate applications from one VM to another. 3. Migrate one VM from one PM to another or create new VM on appropriate PM. (BIP problem, NP-hard) 4. Turn on/off PM. 5. Outsource to other Cloud 15 provider.
How to evaluate KM techniques? 16
How to evaluate KM techniques? " # $!!! 17
How to evaluate KM techniques? 18
How to evaluate KM techniques? 19
Escalation Level 1: VM reconfiguration Speculative approach: May we allocate less resources then agreed, but more than actually utilized at the specific point in time and not violate SLAs? What do we provide? What does the consumer utilize? Storage 20 What was agreed in the SLA? 500 GB 400 GB >= 1000 GB NO 500 GB 510 GB >= 1000 GB YES 1000 GB 1010 GB >= 1000 GB NO Violation?
Methodology I Case Based Reasoning (CBR), Rule-based approach Threat Thresholds Self-adaptive 21
Evaluation Evaluation )"!#$!"(#$!"'#$!"&#$!"%#$!"!#$ )$ %$ *$ &$ +$ '$,$ 2 3 4 5 6 7 8 TTlow 30% 30% 30% 50% 50% 50% 70% 70% TThigh 60% 75% 90% 60% 75% 90% 75% 90% ("#$ '!#$ '"#$ &!#$ &"#$ %!#$ %"#$!!#$!"#$ '!"#!"#$%&'()*' )"%#$!"#$%&"'()*+,)!"#$%&#'()*+,) )"&#$ 1 *$ +$ "##$!'#$!&#$!%#$!"#$ ($ %$ )$,-./012"%,$!$ %$ &$ '$ (# &$ *$ '$ $# )# %# *# &# +# '# +",%-./$' +!!" *!!" )!!" (!!" '!!" &!!" %!!" $!!" #!!"!"!#!!&$"!#!!&"!"#$%&#'(% ""#$!"#$!"#$%&!'"()%*+% "%#$ "$ $!"# -./(&0$') -./'%0"#)!$ %!"#!"# )$ ($ &!"#!#!!%$"!#!!%"!#!!!$"!" #" $" %" &" '" %&'()*+,$ 22 (" )" *"!" %!!!" &!!!" )*'% '!!!"
Methodology II Power consumption, migration and power management models Heuristics for VM migration and PM power management First placement of VMs Re-allocations of VMs 23
Migration t t+1 VM 1 PM1 PM2 Decision: Migrate VM1 from PM1 to PM2! t+2 VM 1 VM 1 PM1 PM2 VM 1 PM1 PM2
Powering on PM t t+1 PM1 Decision: Power on PM1! not available for VMs, but consuming full energy level (Cmax) t+2 PM1 Ready for allocation of VMs, but able to run VM only at t+3 (cf. migration)
Powering off PM t PM1 Decision: Power off PM1! Precondition: No VMs are executing on PM1 t+1 t+2 PM1 not available for VMs, but consuming full energy level (Cmax) PM1 powered off
Evaluation 27
Evaluation 28
Evaluation 29
Evaluation 30
Evaluation 31
Conclusion and Outlook Modeling and simulating a Cloud infrastructure Enacting SLAs and reducing energy consumption on different levels: VMs, PMs, migrations, etc. Realizing energy efficiency not only as best effort, but as strict energy cap Implementation on a real system 1. Small internal cluster 2. Manage a part of VSC2? 32
Michael Maurer Distributed Systems Group Institute of Information Systems Vienna University of Technology Austria maurer@infosys.tuwien.ac.at http://www.infosys.tuwien.ac.at/staff/maurer/