Towards Sustainable Networks
Energy Efficiency Policy from Business to Device Instance Levels
Tereza C. M. B. Carvalho
1
, Ana C. Riekstin
1
, Marcelo C. Amaral
1
, Carlos H. A. Costa
1
,
Guilherme C. Januario
1
, Cristina K. Dominicini
1
and Catalin Meirosu
2
1
Escola Polit
´
ecnica, University of S
˜
ao Paulo, S
˜
ao Paulo, Brazil
2
Ericsson Research, Packet Technologies, Stockholm, Sweden
Keywords:
Sustainability Policy-based Network Management System, Policy Refinement, Policy-based Network
Management.
Abstract:
Policy-Based Systems have been proposed as a flexible and robust solution for policy enforcement in differ-
ent sectors of ICT (Information and Communication Technology) infrastructure. Since networking systems
present high degrees of electrical energy consumption, companies that provide ICT services or network con-
nections need to work on energy-efficient networking processes to guarantee sustainable operations. This
paper presents how an energy efficiency–related business policy can be refined to network policies that con-
sider QoS constraints, how these policies manifest themselves in the equipments, and describes research’s next
steps, which involve making this refinement more automated.
1 INTRODUCTION
In 2007, Gartner found out that ICT (Information and
Communication Technology) was responsible in av-
erage for 2% of energy expenses in some countries
(GeSI, 2008). However, this contribution varies from
country to country as discussed in (Bolla et al., 2011).
In some countries, as Germany, France and Japan,
ICT contribution can reach up to 10%. In the total of
energy consumption, typically, the network commu-
nication systems are responsible for 70%, Data Cen-
ters for 20%, and other systems for 10%. In this sce-
nario, when a company has as policy to be energy ef-
ficient, it has to work on energy efficient networking
processes. But, at the same time, if this company pro-
vides ICT services, or more specifically network con-
nections, as does an ISP (Internet Service Provider),
the energy efficiency strategy cannot interfere in the
provided QoS (Quality of Service), normally set in
SLAs (Service Level Agreement).
Considering this, the present research project aims
at developing an energy-efficient policy-based net-
work management system architecture. A policy is
considered as a set of rules
1
used to administer, man-
age and control the access to ICT resources and ser-
vices (Strassner, 2003). A policy can be viewed as a
set of hierarchical levels so-called Policy-Continuum
as the first is the Business; the second is the System;
the third is the Network; the fourth is the Device; and
the fifth is the Instance level (Strassner, 2003).
An optimized energy-efficiency control requires
a centralized network view, measuring the efficiency
and the performance of each node in the network.
One of the research questions inside this arena is how
to translate an Energy Efficiency Policy defined at
the business level to an Energy Efficiency Policy at
network level. It is necessary to consider QoS con-
straints, since SLAs have to be specified in a machine-
readable form (Klingert and Bunse, 2011). The
mapping between SLAs requirements into machine-
readable form is known as “Policy Refinement Prob-
lem” (Rubio-Loyola et al., 2006) and is clearly an
open issue in the business and ICT alignment discus-
sions.
In this way, one attempt is the work of (Rubio-
Loyola et al., 2006) that uses a methodological ap-
proach to do this refinement, but not considering
energy-efficiency aspects. Then, this paper aims at re-
1
Policy rules have four main components: role - the con-
text where the rule is applied; condition - informs under
which state the rule should be applied; priority - in the case
of rules conflict, it defines the priority among them; and
action - describes the procedure to be performed if the con-
dition rule is satisfied.
238
C. M. B. Carvalho T., C. Riekstin A., C. Amaral M., H. A. Costa C., C. Januario G., K. Dominicini C. and Meirosu C..
Towards Sustainable Networks - Energy Efficiency Policy from Business to Device Instance Levels.
DOI: 10.5220/0004097702380243
In Proceedings of the 14th International Conference on Enterprise Information Systems (ICEIS-2012), pages 238-243
ISBN: 978-989-8565-12-9
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
fining an energy efficiency policy decided at business
level down to the instance level to verify how such a
policy can be considered in a network management
system, using a methodological approach. As far as
we know, this work is the first step towards an auto-
mated approach, which is an open issue (Uszok et al.,
2008) and an important way to bind business and net-
work policies. By now, the progress in this field is
slow (Han and Lei, 2012).
For such discussion, in Section 2, the basic archi-
tecture for Policy-Based Systems and the policy lev-
els of an ICT system are presented. Section 3 shows
how policies are described. The mapping between en-
ergy efficiency policys levels of abstraction is shown
in Section 4. At last, Section 5 summarizes this paper
and presents the next steps.
2 POLICY-BASED
MANAGEMENT
Policy-Based Management (PBM) uses system poli-
cies to control the overall system (Strassner, 2003). In
the case of networks, it is called PBNM, Policy-Based
Network Management. The architecture for such a
system is described in Figure 1.
The operation model for this architecture consists
of: policies creation, modification and storage, per-
formed by the Policy Management System (PMS);
search and retrieval of stored policies, by the Policy
Decision Point (PDP); and policies enforcement in
ICT processes or resources, by the Policy Enforce-
ment Point (PEP).
Policy Management
System
Dedicated
Policy
Repository
Policy Decision
Point
Policy Enforcement
Point
Events
Figure 1: Policy-based network management.
According to the Internet Engineering Task Force
(IETF), in ICT systems, policies can have different
levels of abstraction with different scopes, such as il-
lustrated in Figure 2 (Strassner, 2003). These scopes
can be divided into “ICT Governance”, responsible
for the Policy Enforcement at Business and System
levels, and “Network Management System”, respon-
sible for the Policy Enforcement at Network, Device
and Instance levels.
SLA, Processes, Guides and
Goals
Variables applied to Devices and
its Components (e.g. MIB)
Sustainability and Performance
Indicators
Metrics for Network Operations
related to its Technology
Metrics for Device Operation
Network
Management
System
Business View
System View
Network View
Device View
Instance View
ICT
Governance
Figure 2: Policy levels.
A policy for network management can be speci-
fied at the business level, the first layer, through an
SLA (Service Level Agreement). This policy has to
be translated to the system level, the second layer,
by using performance indicators. At the network
level, the third layer, policies are described making
use of structured languages, without the details about
the network devices. As examples of structured lan-
guages deployed at the network policy level, Ponder
(Lymberopoulos et al., 2003), SWRL (Horrocks et al.,
2004), REWERSE (Bonatti et al., 2005) can be men-
tioned. The fourth layer, device level, shows how this
information is used to reconfigure the equipments.
The fifth layer, instance level, shows the specific con-
figuration commands on a per-device and vendor ba-
sis. Typically, policies belonging to a policy contin-
uum differ among themselves in syntax aspects, but
they are related through semantics aspects. The trans-
lation between all these layers is not easy, and au-
tomated approaches are under development, such as
KAoS (Uszok et al., 2008). This issue, known as
“Policy Refinement Problem” (Rubio-Loyola et al.,
2006) is one of this research’s next steps.
3 POLICY SPECIFICATION
This section discusses how the policies described in
Section 2 are specified.
3.1 Business and System Policies
The policies at the business and system levels are nor-
mally specified using natural language. Each pol-
icy defines its goals, conditions, associated indica-
tors, actions, and scope. These policies have to be re-
fined down to a machine-readable form (Klingert and
Bunse, 2011). KAoS, for instance, uses constrained
English sentences (Van der Meer et al., 2006) that are
refined down to the network level, but, according to
(Uszok et al., 2008), the automated approach is an
open issue. (Han and Lei, 2012) say that the progress
TowardsSustainableNetworks-EnergyEfficiencyPolicyfromBusinesstoDeviceInstanceLevels
239
in this field is slow. It is important to develop this
automated approach, once one consider that the com-
panies’ managers (business level) use a different lan-
guage than the one used by the ICT/network managers
(Strassner, 2003) and that, solely with the usual policy
structure ’if, then, else’, it could be difficult for them
to define business policies.
3.2 Network Policy
Among the structured languages used to describe
network policies, such as SWRL (Horrocks et al.,
2004) and REWERSE (Bonatti et al., 2005), Ponder
is a flexible and well accepted language (Rana and
Foghlu, 2009). It is a specification language for man-
agement and security policies applied onto distributed
systems. It was developed by a research group on
Policies for Distributed Systems from London Impe-
rial College. For policies specification, analysis and
execution, a Ponder tool set can be composed by a
compiler, an editor, and a management kit for sup-
porting management of the model’s modules from an
administrative console.
3.3 Device and Instance Policy
The policy at device level is defined by a set of device
commands. Thus, it is necessary to translate a pol-
icy described by a declarative language, as Ponder,
to a policy described as a set of device commands.
The challenge involves the translation from abstract
policies to concrete policies. For correlating concepts
from two different policy languages, it is necessary
to define a semantic model that specifies the vocabu-
lary’s semantics and concepts’ definitions adopted by
the policy language. This purpose is achieved by in-
formation models such as CIM - Common Informa-
tion Model; SID - Shared Information/Data Model;
DEN - Directory Enabled Networks, and the next gen-
eration version, DEN-ng, which defines the mapping
from abstract policies to concrete ones. This common
semantics is used in the translation from the abstract
policy to the concrete policy, making it understand-
able by devices. The concrete policy results in device-
specific commands.
The difference between the device and the in-
stance levels can be better understood by an exam-
ple of a policy related to energy efficiency. On one
hand, the device level should specify in an abstract
way which functionalities and protocols are available
for each device in order to expose how the policy rules
will be represented in these devices. An example is
the description of the supported power modes, such as
the “sleep” or the “powered on” or even other modes
and how it will be handled, such as using SNMP along
with a specific MIB. On the other hand, the instance
level may be composed by command line for the net-
work administrator to apply actions into the devices.
An example is a set of SNMP commands to set the
device to “low-energy-consumption” mode, such as
“state 7” that means the “sleep mode” in the EMAN-
MONITORING-MIB (Chandramouli et al., 2011). A
complete example of how a policy can be refined
through all policy levels is described in the next sec-
tion.
4 SUSTAINABILITY-ORIENTED
POLICY AND ISSUES
This section shows how a sustainability policy defined
at business level can impact and be mapped in differ-
ent policies in all levels, down to the instance one.
At the business level, it is specified a policy that can
cover the whole company. The policy is related to
“Sustainable Practices”, as illustrated in Figure 3. As
an example of “Sustainable Practice” it was selected
“Rational Use of Resources”.
Business
ICT
Devices
Network
Network
Data
Center
Personal
Computing
Lighting
Air
Conditioning
DevicesDevices
DeviceInstance
DevicesDevicesInstances
Business
Sustainable
Practices
Rational Use
Electrical
Power
Water
Diesel
Oil
System
of Resources
Figure 3: Policy mapping.
At the System Level, “Rational Use of Resources”
involves rational use of water, diesel oil and electri-
cal power, among other resources. As a matter of ex-
ample, it was chosen “electrical power”. At the In-
frastructure Level, the rational use of electrical power
includes mainly lighting, ICT (Information and Com-
munication Technology) and air conditioning. As ICT
infrastructure, it can be considered mainly Data Cen-
ter, Network and Personal Computing. In this case,
“Network”, as an important energy consumer, was se-
lected for further details. At the Device Level, one
must consider that a network is composed by sev-
eral devices (e.g., routers, switches, line regenerators)
connected by communication links. If the goal is to
ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems
240
have an energy efficient network, it will impact all
of its devices. At last, at the Instance Level, any en-
ergy efficiency operation on some device will affect
its components, as interfaces and ports. Such impact
on device’s components is analyzed at the Instance
level.
At the first two layers, Business and System, poli-
cies are described using natural language. This high-
level policy description is translated to the Network
Policy level - in this example, using Ponder language.
From this translated description, models that take into
account the supported functionalities of each device
are described. This generates the device level de-
scription. Some of these functionalities are described
through subsets of MIB (Management Information
Base) objects. These objects can include device’s de-
mand and performance, energy consumption, etc. At
last, the device description is translated into instance
description, a process which regards, for each specific
devices, SNMP operations based on those MIB ob-
jects.
This policy mapping that takes place between dif-
ferent policy levels is presented and discussed in
much more details in the next sections. There, it is
presented a real example of an energy-efficiency busi-
ness policy and, using a methodological approach, it
is shown how to refine it down to the instance level.
The goal is to verify that it is possible to do the policy
refinement using existent tools to, in the near future,
work in automated approaches, the main goal of this
research.
4.1 Business and System Policies
As example of business policy, was taken the follow-
ing policy: “All the ICT (Information and Communi-
cation Technology) processes and procedures should
be aligned with sustainable practices”. This policy is
associated to several business goals. In this case, it
was considered the goal related to “Resources Ratio-
nal Use”, which naturally includes energy efficiency.
From this policy at the business level, one or more
system-level policies are generated. Among differ-
ent policies related to energy efficiency, one example
is: “deactivation of network equipments that are idle
more than 70% of the time, if it is possible to redi-
rect their traffic and keep the network QoS (Quality
of Service) negotiated with the customer (established
in SLA)”. To reach the performance requirements, the
following rule could be applied: “keep active network
equipments with use rate greater than 30%”.
4.2 Network Policy
The translation of the aforementioned System Pol-
icy, for instance, can result in the following Network
Level policies (one policy at system level can be trans-
lated into one or more in network level):
• Policy 1: “If some network equipment is idle more
than 70% of the time and the network delay is less
than 60 milliseconds, deactivate this equipment if
its traffic load can be redirected to other equip-
ment”, i.e., if there is any redundant path. This
policy consists on the identification of underuti-
lization metrics and the subsequential optimiza-
tion of the resources usage as long as there would
be no degradation of the provided QoS (e.g., the
network delay increase). The purpose is to get the
best usage of the redundant paths that typically
exists for QoS maintenance.
• Policy 2: “If the network delay is greater than
80 milliseconds and there are redundant paths
with deactivated equipments, then reactivate these
equipments”. This policy has as purpose activa-
tion of equipments that had been deactivated due
to some other policy actions.
Figure 4 shows the description of Policy 1
using Ponder language. The Block 1 defines the
limit operation value for the network parame-
ter used in this example: the router minimum
usage rate (30%). The Block 2 defines an au-
thorization policy, i.e., an access control pol-
icy authorizing the service management agent
(/PolicyAgents/ServiceManagementAgent) to
set a router, belonging to the group /Routers, to the
sleep mode if it is possible to transfer the traffic of this
router to another one. This restriction is defined by the
function isPossibleToTransferTrafficLoad(r)
and avoids the network graph disruption.
The Block 3 defines an obligation policy
(/Policies/LowUtilizationPolicy), i.e., a
policy that executes an action on an object when a
specified event happens. This policy makes the agent
(/PolicyAgents/ServiceManagementAgent) to
set a router, belonging to the group /Routers/, to
the sleep mode, when an event arrives informing the
low utilization of such router.
4.3 Device Policy
Policies at the device level were represented by a
model that describes the functionalities that each de-
vice has to provide in order to apply the Network Pol-
icy. This level is directly related to the device’s type
and vendor, since same types of devices from different
TowardsSustainableNetworks-EnergyEfficiencyPolicyfromBusinesstoDeviceInstanceLevels
241
// Block 1:
minUtilization = 0.3;
// Block 2:
root at: "authorizationdomain" put:
root/factory/domain create.
newauthorizationpolicy := root load:
"AuthorisationPolicy".
root/factory at: "newauthorizationpolicy" put:
newauthorizationpolicy.
root/authorizationdomain at:
"DisableRouterAuthPolicy" put:
(newauthorizationpolicy subject:
root/PolicyAgents/ServiceManagementAgent
action: setSleepMode
target: root/Routers
when isPossibleToTransferTrafficLoad).
root/authorizationdomain/DisableRouterAuthPoli
cy
active: true.
// Block 3:
LowUtilizationPolicy :=
root/factory/ecapolicy create.
LowUtilizationPolicy event:=
root/event/UtilizationValue.
LowUtilizationPolicy condition:
[:oldUtilization :newUtilization |
(newUtilization < minUtilization)
& (oldUtilization >= minUtilization)].
LowUtilizationPolicy action: [
root/Routers setSleepMode: true; show ].
LowUtilizationPolicy active: true.
Figure 4: Example of network policy description in Ponder
- Policy 1.
//Example - policy 1
Not dependant on Operating System
functionalities;
SNMP protocol support
MIB support: RFC1213-MIB, TN3270E-RT-MIB,
POWER-MONITOR-MIB
Power states support: Sleep, Low e High
Figure 5: Example of device level policy.
vendors usually provide different functionalities. In
this example, it is adopted the hypothetical equipment
model described in Figure 5. The actions described
will take place at the device level of this equipment.
The Figure 5 model represents a Device Policy
and has the following information: mapping how the
device implements the non-standardized functionali-
ties. In this example, it is necessary to map the fol-
lowing power modes (Chandramouli et al., 2011):
1. SleepMode: the device’s awakening functional-
ity is available and the power consumption is
near zero, corresponding to G1 and S3 levels in
the Advanced Configuration and Power Interface
(ACPI), an open industry specification that en-
ables power management of mobile, desktop, and
snmpget [options] [-Cf] [OID]
Options: protocol version, device address,
MIB name
-Cf : try to correct errors
OID: variables to be read
snmpset [options] [-Cf] [OID] [type] [value]
Options: protocol version, device address,
MIB name
-Cf : try to correct errors
OID: variables to be set
type: i (integer), t (time), a (ip address),
s (string), D (floating point), ...
Figure 6: SNMP commands’ syntax.
server platforms (Steele, 1998).
2. Low: indicates that essential functionalities are
available and power consumption is reduced, cor-
responding to G0, S0 and P4 levels in the ACPI.
3. High: represents that all device functionalities are
available and the power consumption is greater
than the other modes, corresponding to G0, S0
and P0 levels in the ACPI.
In this example, the device supports SNMP (Sim-
ple Network Management Protocol) and the following
MIB (Management Information Base) object sets:
1. RFC1213-MIB: provides data for bandwidth uti-
lization calculation, as the total number of ingo-
ing (ifInOctets) and outgoing (ifOutOctets)
bytes in the device and total transmission capac-
ity (ifSpeed). The throughput (T ) in Mbps, can
be obtained as follows:
T =
max(∆ifInOctets, ∆ifOutOctets) × 8 ×100
(seconds in ∆)ifSpeed
2. TN3270E-RT-MIB: provides data related to the
network delay and average reply time since last
measured time interval, etc.
3. POWER-MONITOR-MIB (Chandramouli et al.,
2011): provides the device power state (Power-
MonitorLevel), which varies from 1 to 12.
4.4 Instance Policy
Instance Level Policies describe how the function-
alities at the device level are implemented via spe-
cific instance’s language commands. In the case of
this example, the snmpget and snmpset commands
of the SNMP protocol are used to translate the in-
stance level policy. The snmpget is used to retrieve
information from network devices, providing one or
more variables (object indentifiers, or OIDs) as pa-
rameters. The snmpset is used for information mod-
ification at a remote device and has similar syntax to
ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems
242
// Example "Policy 1"
//Collecting information for calcula
ting the band-width utilization:
snmpget -v 2c -c public target_ip_address
-m RFC1213-MIB ifInOctets.0
snmpget -v 2c -c public target_ip_address
-m RFC1213-MIB ifOutOctets.0
snmpget -v 2c -c public target_ip_address
-m RFC1213-MIB ifSpeed.0
//Collecting information about
the network delay
snmpget -v 2c -c public target_ip_address
-m TN3270E-RT-MIB
tn3270eRtDataAvgRt.0
// Configuring the CPU power state
snmpset -v 2c -c public target_ip_address
-m POWER-MONITOR-MIB
PowerMonitorLevel i 4
Figure 7: Example of Policy 1 translation to SNMP com-
mands at the Instance level.
snmpget’s syntax with an additional parameter related
to the value to be set. Figure 6 describes these com-
mands’ syntaxes.
The Figure 7 presents the example of Policy 1
translation to SNMP commands at the Instance Pol-
icy level.
5 FINAL CONSIDERATIONS
Considering the increasing demand for ICT systems,
especially for networking services, it is even more im-
portant to set energy efficiency policies for these ser-
vices as basis for supporting Carbon free systems.
This paper presented how to refine an energy ef-
ficiency business policy down to the instance level,
considering the Policy Continuum’s five levels: Busi-
ness, System, Network, Device and Instance. The first
step was to use a methodological approach to prove
that it is possible to consider an energy-efficient pol-
icy in a network management system. The next step
is to develop an automated approach for the “Policy
Refinement Problem”.
This automated approach is important since the
business managers will be able to define business
policies that will be directly applied to the network,
devices and instances, making the business–IT align-
ment easier.
ACKNOWLEDGEMENTS
This work was supported by the Innovation Center,
Ericsson Telecomunicac¸
˜
oes S.A., Brazil.
REFERENCES
Bolla, R., Bruschi, R., Davoli, F., and Cucchietti, F. (2011).
Energy efficiency in the future internet: A survey of
existing approaches and trends in energy-aware fixed
network infrastructures. Communications Surveys Tu-
torials, IEEE, 13(2):223 –244.
Bonatti, P. A., Antoniou, G., Baldoni, M., Baroglio,
C., Duma, C., Fuchs, N., Martelli, A., Nejdl, W.,
Olmedilla, D., Peer, J., Patti, V., and Shamheri, N.
(2005). The rewerse view on policies. In ISWC Se-
mantic Web Policy Workshop (SWPW).
Chandramouli, M., Schoening, B., Quittek, J., Dietz, T., and
Claise, B. (2011). Power and energy monitoring mib.
Technical report, Network Working Group, IETF.
GeSI (2008). Smart 2020: Enabling the low carbon econ-
omy in the information age. Technical report, The
Climate Group on behalf of the Global eSustainability
Initiative (GeSI).
Han, W. and Lei, C. (2012). A survey on policy languages
in network and security management. Computer Net-
works, 56(1):477 – 489.
Horrocks, I., Patel-Schneider, P., Boley, H., Tabet, S.,
Grosof, B., and Mike, D. (2004). SWRL: A semantic
web rule language combining owl and ruleml. Tech-
nical report, W3C.
Klingert, S. and Bunse, C. (2011). Greenslas for energy effi-
cient scheduling. In Proceedings of the COST Action
IC0804 - 2nd Year, COST IC0804, pages 122–126.
European Cooperation in Science and Technology.
Lymberopoulos, L., Lupu, E., and Sloman, M. (2003). An
adaptive policy-based framework for network services
management. J. Netw. Syst. Manage., 11(3):277–303.
Rana, A. and Foghlu, M. O. (2009). New role of policy-
based management in home area networks - concepts,
constraints and challenges. In New Technologies, Mo-
bility and Security (NTMS), pages 1 –6.
Rubio-Loyola, J., Serrat, J., Charalambides, M., Flegkas, P.,
and Pavlou, G. (2006). A methodological approach
toward the refinement problem in policy-based man-
agement systems. Communications Magazine, IEEE,
44(10):60 –68.
Steele, J. (1998). ACPI thermal sensing and control in the
PC. In Wescon/98, pages 169 –182.
Strassner, J. (2003). Policy-Based Network Management:
Solutions for the Next Generation. Morgan Kauf-
mann, 1 edition.
Uszok, A., Bradshaw, J. M., Lott, J., Breedy, M., Bunch, L.,
Feltovich, P., Johnson, M., and Jung, H. (2008). New
developments in ontology-based policy management:
Increasing the practicality and comprehensiveness of
kaos. In IEEE Workshop on Policies for Distributed
Systems and Networks, POLICY ’08, pages 145–152,
Washington, DC, USA. IEEE Computer Society.
Van der Meer, S., Davy, A., Davy, S., Carroll, R., Jennings,
B., and Strassner, J. (2006). Autonomic networking:
Prototype implementation of the policy continuum. In
Broadband Convergence Networks, 2006. BcN 2006.
The 1st International Workshop on, pages 1 –10.
TowardsSustainableNetworks-EnergyEfficiencyPolicyfromBusinesstoDeviceInstanceLevels
243
