GreenSLAs

Providing Energy Consumption Flexibility in DCs through Energy-aware

Contracts

Juan Felipe Botero

, Sonja Klingert

, Xavier Hesselbach-Serra

Antonella Falcone

and Giovanni Guiliani

Universitat Politecnica de Catalunya, Barcelona, Spain

University of Mannheim, Mannheim, Germany

HP Italy Innovation Centre, Milano, Italy

Keywords: Energy Consumption, Energy Saving, DCs, GreenSLA, Collaboration Level, Flexibility.

Abstract: This paper describes how eco-motivated extensions added to the traditional SLA concept can provide levels

of flexibility in the data centre management that enable the data centre to reduce the energy consumption,

improve the environmental footprint and lower internal costs. A close collaboration with both the IT

customer and the data centre’s energy provider are key to this approach which is formalized by means of the

GreenSLA concept building on the three main components: Flexibility, GreenKPIs and collaboration.

Performance decisions regarding the energy consumed by the services are taken and translated to actions to

be executed to save energy while at the same time guaranteeing the contract agreements.

1 INTRODUCTION

Electricity consumed in DCs, including enterprise

ICT equipment, air cooling devices (AC) and

uninterruptable power supply systems (UPS), is

expected to contribute substantially to the electricity

consumed in the European Union (EU) commercial

sector in the near future, especially with the cloud

computing trend still on the rise.

Figure 1: IT Service.

The project All4Green analyses the relationship

between the IT Customer (ITC) and DC (DC) or DC

federation in conjunction with the relationship

between the DC and the EP (EP), aiming to unlock

the untapped potential for globally saving energy

and CO2 emissions by proposing enhanced

collaboration mechanisms within the DC ecosystem.

New flexible contracts (Green-SLAs) between IT

users and DCs are used to enable efficient energy

saving policies tailored to different computing styles

and to enable a better matching of DCs energy

demand profiles with energy supply patterns of the

EPs.

2 GreenSLAs IN ALL4GREEN

The liabilities between a DC and its customers are

ruled by a set of contracts. Apart from framework

contracts, for the delivery of each DC service a

service level agreement (SLA) is closed. In the

context of this paper we state that in a DC an

arbitrary collection of servers, software and VM

entities jointly provide an IT service to end-users as

shown in figure 1.

The SLA that is connected to each service

contains at least a subset of the following elements:

119

Felipe Botero J., Klingert S., Hesselbach-Serra X., Falcone A. and Giuliani G..

GreenSLAs - Providing Energy Consumption Flexibility in DCs through Energy-aware Contracts.

DOI: 10.5220/0004379401190122

In Proceedings of the 2nd International Conference on Smart Grids and Green IT Systems (SMARTGREENS-2013), pages 119-122

ISBN: 978-989-8565-55-6

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

A functional description of the service, performance

requirements, availability, maintenance work plan,

specific execution requirements, financial penalties

and rewards.

As DCs deliver an abundance of IT services with

as many SLAs to be monitored and executed, the

formal WS-Agreement framework has become a de

facto standard in research projects due to its

flexibility in integrating agreement-specific

elements.

The WS-Agreement is denoted in XML and is

made up of the following building blocks as

depicted in figure 2: Name, context, and terms.

Within the terms, service terms are definition and

description elements whereas guarantee terms

provide metrics based on the definitions in the

service terms and indicate the contracted boundaries

for these metrics.

Figure 2: WS-Agreements.

Until now, however, SLAs were very static in nature

and restricted to a very function point of view. As

the environmental impact of IT services all over the

world needs to be reduced, the idea of an eco-

motivated SLA, a so-called GreenSLA was

developed.

3 GreenSLA

Previous research has been made in the context of

GreenSLA (Laszewski and Wang, 2010), (Klingert

et al., 2011) considering the DC as an atomic unit.

However, All4Green goes one step further by

including the main actors of the DC’s ecosystem

with EP and ITC. In All4Green, GreenSLAs are

extensions to traditional SLAs including three main

additions:

 Flexibility,

 GreenKPIs and

 Collaboration

The flexibility is the variability that the ITC and

the DC are willing to accept in each of the service’s

running conditions requirements (e.g. performance/

availability/execution/maintenance) due to a change

in the context. The context can be defined as the

specific situation that provokes the possibility of

modifying the service conditions to promote a more

environmentally friendly behaviour. The time span

and context dependency of these situations should

lead to the reduction of energy consumption or CO

emissions while executing a service. For instance, a

time-dependent IT service can work at high

performance level from 8am to 10pm and decrease

to low performance from 10pm to 8am.

Similarly, green KPIs are new service level

objectives also pursuing the goals for energy

consumption and/or CO

emission reduction. For

instance, a GreenSLA could guarantee a certain CO

emissions level during the execution of an IT service

by allowing continuous interaction between the DC

and the EP.

Collaboration in the DC-ITC sub-ecosystem is

regulated by the GreenSLA and refers to the

requests to the ITC coming from the DC. These

requests are triggered by the change of the DC

mode: From regular mode the DC might switch to

situations where it needs to reduce its current energy

consumption (energy saving mode) or, on the

contrary, switch to a situation where it needs to

consume extra available energy (extra energy mode).

This change of DC status comes from the interaction

of the actors in the DC-EP sub-ecosystem.

3.1 Flexibility

Flexibility will depend on a predetermined context.

This context is the situation and the condition in

which the guaranteed level of a determined service

item can be modified in order to effect energy and/or

emissions savings. GreenSLAs differentiate

between two different context types: time/calendar-

dependency and energy DC mode-dependency.

Time/Calendar-dependency: This context is

related to a time period: hour(s) and/or day(s) of the

week and/or the year’s season. As mentioned above,

an example of a time-dependent GreenSLA clause

could be: High Availability + High Performance in

working days, and low availability + low

performance during nights and weekends.

DC Energy Mode-dependency: As shown in

figure 3, this context is related to the current energy

mode of the DC. Generally, the DC is in Regular

Energy mode (RE), but it can change to Extra

Energy (EE) mode or to Energy Saving (ES) mode

for a certain time period due to the reception and

SMARTGREENS2013-2ndInternationalConferenceonSmartGridsandGreenITSystems

120

acceptance of a request coming from the EP.

Figure 3: Possible Datacentre States.

In ES mode the DC attempts to save energy by,

amongst others, downgrading IT services and/or

shifting tasks in time, whereas in EE mode the DC

could promote IT services to higher performance

levels, or anticipate the execution of maintenance

tasks. GreenSLAs should offer flexible clauses to

regulate and support this behaviour. For example, an

energy mode-dependent GreenSLA clause could be:

Medium performance + medium availability in

regular mode and Low performance + Low

availability in ES mode.

3.2 GreenKPIs

GreenKPIs have the double function of monitoring

the eco-sustainability of a service and giving

feedback to the ITC that the GreenSLA guarantees a

win-win situation not only from the economic but

also from the environmental point of view.

Examples for GreenKPIs can be a guaranteed

energy mix for the DC or a boundary for CO2

emissions generated or KWh consumed by a certain

service.

3.3 Collaboration

The Collaboration level between DC and ITC is

tightly (but not exclusively) bound to the existing

collaboration agreements between EP and DC. This

means that the DC knows its degree of commitment

with the EP before setting up GreenSLAs with its

ITCs.

A collaboration request sent from the DC to the

ITC is generally originated by the EP-DC

collaboration. As shown in figure 4 the EP can

request a reduction of energy consumption or may

demand higher energy consumption due to a surplus

of renewable energy. In order to accept this request,

the DC may initiate a negotiation with the ITC by

requesting the permit to execute a set of actions

that modify the current behavior of the IT service.

Figure 4: Cloud GreenSLA template.

An example is to ask the ITC if a VM can be paused

for e.g. 2 hours – the response very likely depends

on the current status of the application running

inside the VM (that is known to the ITC, but not to

the DC).

Upon the reception of a collaboration request, the

ITC has the freedom to reject but also the obligation

to accept a certain number of times. Otherwise it

loses its right to an economical reward. GreenSLAs

also regulates the degree of freedom of the ITC

Agent and the voracity of the DC.

3.4 Pricing

It is the basic principle of GreenSLAs that the ITC

renounces pre-specified performance aspects in

order to receive a reward in return. This is the

incentive aspect of GreenSLAs which was

described, e.g. in (Klingert et al., 2011), (Bunse et

al., 2012). The incentive can be either financial or

non-financial like e.g. a special treatment for green

customers, or a combination of both. In All4Green, a

purely financial incentive was chosen, as most of the

use cases involved a B2B-relationship between DC

and ITC.

Generally, the pricing of an IT service depends

on a fix basic fee and variable components, like

transaction fees and prices per service instance.

To fulfil the incentive function the expected

GreenSLA reward must to a great degree surpass the

expected GreenSLA penalty. The reward in

All4Green depends on the (contracted) flexibility

that the IT customer agreed on in the GreenSLA, the

(measured) QoS degradation the ITC has to endure,

and the (measured) collaboration effort between the

ITC and DC. The contracted (ex-ante) part of the

reward by nature is static in the way that it composes

an unchangeable (and thus reliable) part of the

reward as long as the contract, the GreenSLA, is

GreenSLAs-ProvidingEnergyConsumptionFlexibilityinDCsthroughEnergy-awareContracts

121

valid. The measured (ex-post) parts of the reward by

nature are dynamic in the way that they change

along with the behaviour of the customer as well as

the technical implications of the activity of the DC

management within the All4Green framework. Thus

the reward formula can be expressed as:

GreenSLAReward=StaticReward(Flexibility)+D

ynamicReward(ΔQoS,realizedCollaboration)-

Penalty(realizedCollaboration)

Penalties are imposed in case the ITC’s

collaboration is lower than contractually agreed.

4 EXAMPLE

This section presents an example of GreenSLA for a

simple IT service offered by a cloud provider. Cloud

computing services mainly consist of VMs and

include KPI detailed parameters such as Boot delay,

computing power, memory, OS image, I/O

performance or disk size. For the sake of simplicity,

a subset of the offered service portfolio is provided:

 VMs have “sizes” composed of a pre-determined

number of virtual CPUs (VCPUs), RAM and HD

capacity.

 VMs run “images” of Operating Systems.

 The Performance KPI limits in equivalent

percentage of CPU units (ECUs).

 The Boot delay KPI limits the start-up time.

A template that generates a GreenSLA for the

Green VM service as described in this section is

shown in figure 4: It represents the renting of a VM

with a large size and Windows 2008 and the

following specific arrangements for the GreenSLA

parameters:

As stated in the previous section, the flexibility

depends on the time/calendar and DC energy mode.

In this example, the time/calendar-dependency

makes evokes very flexible scheduling.

Additionally, the DC energy mode dependency

enables a flexible boot delay and performance.

Concerning a GreenKPI, in this example, the

CO2 emissions, retrieved through the EP of the DC,

are limited: i.e. the energy provided by EP to DC

should be produced at <300g CO2/Kwh. If this

guarantee is breached, a penalty must be paid by the

DC. Concerning collaboration additions, this

GreenSLA allows requests coming from the DC that

ask to pause the VM for a maximum of two hours.

The collaboration limitations in this example are

defined as follows: The DC can issue a maximum of

10 collaboration requests per month to the ITC, and

the ITC can reject a maximum of 5 collaboration

requests per month.

5 EXPECTED ENERGY SAVINGS

The energy savings in the DC due to the GreenSLA

contractual terms related to the collaboration with

customers are expected to be approximately 10%

according to the estimations and trials done so far in

the All4Green project. The EP will be able to avoid

dangerous peaks, thus avoiding large costs and

reducing emissions. These cost savings will be

partially used to compensate the collaborating data

centers, and in the end this will generate a positive

side effect for the ecosystem, both from

Energy/emissions and economic point of view.

6 CONCLUSIONS

This paper has presented the three main components

of the GreenSLA concept (flexibility, GreenKPIs

and Collaboration) and how the DC takes

performance decisions regarding the energy

consumed by the services to save energy compared

to ecosystems that not include these concepts, while

guaranteeing the contract agreements. All4Green

implementation details and a practical example have

been provided.

ACKNOWLEDGEMENTS

This work has been supported by the European FP7

All4Green project (Grant agreement No. 288674)

and the Spanish Government, MICINN, under

research grant TIN2010-20136-C03.

REFERENCES

C. Bunse, S. Klingert, T. Schulze “GreenSLAs:

Supporting Energy Efficiency through Contracts”. In:

“Energy Efficient DCs”, ed. by J.Huusko, H. de Meer,

S.Klingert, A.Somov, Springer LNCS, 2012

S. Klingert, T. Schulze and C.Bunse “GreenSLAs for the

Energy-efficient Management of DCs”. 2nd

International Conference on Energy-Efficient

Computing and Networking, Columbia University,

New York, USA, 2011

G. Laszewski and L. Wang. GreenIT SLAs. Grids and

Service-Oriented Architectures for Service Level

Agreements, pages 77–88, 2010.

SMARTGREENS2013-2ndInternationalConferenceonSmartGridsandGreenITSystems

122