green cloud computing

36
University of St Andrews School of Computer Science 1 Energy Aware Clouds James W. Smith [email protected]

Upload: university-of-st-andrews

Post on 15-Nov-2014

3.260 views

Category:

Technology


0 download

DESCRIPTION

On June 24th I presented to the Dependable Systems Engineering group here in the School of Computer Science, St Andrews. The group meets once a month for a presentation from one of its members over lunch. The presenter talks about their current research, providing a good opportunity to keep up to date with other work within the group.On June 24th I presented to the Dependable Systems Engineering group here in the School of Computer Science, St Andrews. The group meets once a month for a presentation from one of its members over lunch. The presenter talks about their current research, providing a good opportunity to keep up to date with other work within the group.

TRANSCRIPT

Page 1: Green Cloud Computing

University of St Andrews

School of Computer Science

1

Energy Aware Clouds

James W. [email protected]

Page 2: Green Cloud Computing

University of St Andrews

School of Computer Science

2

Introduction• Total Carbon Footprint of the IT industry was 2% of all human activity in

2007

– 830 MtCO2e

– Energy powering devices is 75% of this total

– Need to build sci-fi power or improve efficiency

• IT is beginning to learn that cutting emissions and cutting costs go naturally together

Page 3: Green Cloud Computing

University of St Andrews

School of Computer Science

3

Costs

• Operational costs exceeding purchase costs

• Mainly driven by energy costs

• Even over a relatively short lifespan

Page 4: Green Cloud Computing

University of St Andrews

School of Computer Science

4

so who benefits?

Page 5: Green Cloud Computing

University of St Andrews

School of Computer Science

5

Roadmap• Energy Aware Computing

• Cloud Computing

• Private Clouds

• Virtualisation

• Datacentres

• PUE & Productivity

• Cooling

• Research areas for Energy Efficient Cloud Computing

• Monitoring

• Resource Scaling

• Smart Load Balancing

• Task Consolidation

• Power Efficient Software

• Future Work 5

Page 6: Green Cloud Computing

University of St Andrews

School of Computer Science

6

Energy Aware Computing• Attempting to address problems of energy efficiency in Computing

Systems

– processor chips– cooling

• The overall problem is to “minimise energy used to perform a certain piece of useful work”

– Control resource availability

– Reduce consumption

Page 7: Green Cloud Computing

University of St Andrews

School of Computer Science

7

Page 8: Green Cloud Computing

University of St Andrews

School of Computer Science

8

Green Cloud?

Positive Negative

•Datacentres can become the most efficient centres for computation yet

•Providers will want to increase cost effectiveness

•and be green!

•Datacentres are now consuming 0.5% of all electricity in the world.

•This will only continue to grow!

Page 9: Green Cloud Computing

University of St Andrews

School of Computer Science

9

Private Cloud• Private Cloud Systems have been likened to

• However, Enterprise does have concerns about Cloud systems which Private Clouds can help to address– Security

– Privacy

– Administrative Control

“drinking on your own and calling it a private party” - P Laudenslager, (unknown)

Page 10: Green Cloud Computing

University of St Andrews

School of Computer Science

10

Virtualization

• Virtualization makes clouds run– Run multiple VMs on each physical machine

– Improves utilization, cost effectiveness

• Save Energy– Increase Utilization

– Migrate work?

– Power down unused machines

– Allocated tasks appropriately?

Page 11: Green Cloud Computing

University of St Andrews

School of Computer Science

11

Virtualization (2)

• Performance overhead– intermediate layer

– increased complexity

• Different tasks have different performance costs– for example, using the same physical disk for two or

more VMs...

– and different power consumptions...

Page 12: Green Cloud Computing

University of St Andrews

School of Computer Science

12

Virtualization (3)

• VMs increase utilization, power consumption & heat on a physical machine

• So we need to be careful how much virtualization we do, where we do it and how we prepare for it

• Is it possible to virtualize in an efficient manner?

Page 13: Green Cloud Computing

University of St Andrews

School of Computer Science

13

Page 14: Green Cloud Computing

University of St Andrews

School of Computer Science

14

Page 15: Green Cloud Computing

University of St Andrews

School of Computer Science

15

Is this new?

John McCarthy (1961):

“computation may someday be organised as a public utility”

Page 16: Green Cloud Computing

University of St Andrews

School of Computer Science

16

Datacentres

• The age of the datacentre is here

• One man and a credit card can tap into some of the largest computing resources in the world

Page 17: Green Cloud Computing

University of St Andrews

School of Computer Science

17

Some figures• Datacentres in the USA consume 1.5% of all electricity in

that country

• Energy consumption in this area has doubled in the period 2000-2006

• Only 50% of electricity consumed can be attributed to useful work done by servers, rest goes on cooling, infrastructure etc

United States Environmental Protection Agency (EPA) 2007

Page 18: Green Cloud Computing

University of St Andrews

School of Computer Science

18

Cheap power isn’t always green

• Allow me to be a hippie for a second...

Page 19: Green Cloud Computing

University of St Andrews

School of Computer Science

19

Power Usage Effectiveness

• PUE compares how much energy is used by computing and infrastructure equipment

• Perfect efficiency would give PUE of 1.0

• Most datacentres in the range 1.3 -> 3.0

PUE = Total Facility Power / IT Equipment Power

Page 20: Green Cloud Computing

University of St Andrews

School of Computer Science

20

Datacentre Productivity• PUE is useful but it doesn’t determine productivity over

power

• Step in the Datacentre Productivity Measurement:

• Useful, as EAC likes to think of doing a task for least amount of power

• But how would you measure Useful work?

Datacentre Productivity = Useful Work / Total Facility Power

Page 21: Green Cloud Computing

University of St Andrews

School of Computer Science

21

Cooling• Why do we need to cool?

– Preserve lifetime of components

• Mechanical Engineering– Air or water?

– Direct Heat Exchange

• Computer Science– Smart load balancing?

Page 22: Green Cloud Computing

University of St Andrews

School of Computer Science

22

Research Areas

Page 23: Green Cloud Computing

University of St Andrews

School of Computer Science

23

Monitoring• Reports have estimated that only 13.4% of organisations monitor their

energy consumption!

• Each component in a system must expose their consumption information

• and control mechanisms?

• If such functionality doesn’t exist then 3rd party tool needed

• Yi Yu

• additional complexity

• Software? Hardware?

• A controller can use this information to manage the system

Page 24: Green Cloud Computing

University of St Andrews

School of Computer Science

24

Combining Computation and Cooling

• Traditionally, Cooling & Computation are controlled independently

• Cooling uses CRAC units to cool datacentre to optimum operating temperature

• Computational load is distributed to give best performance

• However, Parolini et al suggest that workload can be distributed smartly according to temperature

• requires unified framework

“Reducing Data Center Energy Consumption via Coordinated Cooling and Load Management” - Parolini, et al 2008

Page 25: Green Cloud Computing

University of St Andrews

School of Computer Science

25

Powering Management• Switch off your lights!!!

• Well, at least migrate your systems between power states

• How much do we switch off?

• Laptop

• sending to sleep still costs energy

• shutting down save more at the cost of additional time

Performance & Response Time vs. Energy Savings

Page 26: Green Cloud Computing

University of St Andrews

School of Computer Science

26

Resource Scaling• Use only the amount of resource required to complete a task

– Give each task a deadline

– Only give resources to allow completion within that deadline

• Speed Scaling– Adjust CPU speed

– Save energy & cooling costs

• Fine for individual components, but how do we do this on a system-wide scale? 2

6

Speed then time and power

Page 27: Green Cloud Computing

University of St Andrews

School of Computer Science

27

Task Consolidation• Keep machines well utilised

• Bin packing problem– Tasks are objects

– Servers are bins

– Resources are dimensions

• Relies upon being able to accurately predict tasks resource requirements– performance adjusting applications?

Page 28: Green Cloud Computing

14

University of St Andrews

School of Computer Science

Load Balancing

• Traditional model– Distribute work evenly

– Each node has equal workload

Page 29: Green Cloud Computing

15

University of St Andrews

School of Computer Science

Load Skewing

• Energy efficient model– “Skew” load

– Give work to nodes while they can handle it

– Power down unused nodes

Page 30: Green Cloud Computing

16

University of St Andrews

School of Computer Science

Power Efficient Software• Different devices consume different amounts of energy doing (roughly)

the same task.

– i.e. Making a call, playing a song

– Why? Difference in hardware & Difference in software implementation

• Is it possible to produce energy efficient software?

– Optimise for time, scalability, robustness, but energy?

Page 31: Green Cloud Computing

31

PES Principles

1. Useful work corresponds to resources consumed

2. Event-based architecture over polling

3. Light on memory

4. Batch I/O requests

Software Modularity?

Page 32: Green Cloud Computing

32

My Work

Page 33: Green Cloud Computing

University of St Andrews

School of Computer Science

33

StACC Private Cloud• So when the StACC cloud works

what does it offer?– a platform for experimentation

• We can control– architecture

– longitivity

– number of nodes

– exact workload

Page 34: Green Cloud Computing

University of St Andrews

School of Computer Science

34

Future Work• Monitor VM performance

• Performance and Energy Consumption

• Write Resource Monitoring Software

• Energy-Smart Control Algorithms for Clouds?

• Based on what? Utilisation? Consumption? Mix?

• Modify Eucalyptus open source software?

Page 35: Green Cloud Computing

University of St Andrews

School of Computer Science

35

Research Question

• Can Cloud Computing have a positive impact on the energy efficiency of IT systems & can private clouds be made more energy efficient?

Page 36: Green Cloud Computing

University of St Andrews

School of Computer Science

36

Questions?