Content uploaded by Rafiullah Khan
Author content
All content in this area was uploaded by Rafiullah Khan on Apr 12, 2017
Content may be subject to copyright.
Smart Proxying for Reducing Network Energy
Consumption
Rafiullah Khan, Raffaele Bolla
DITEN Department
University of Genova (UNIGE)
Via Opera Pia 11a, 16145 Genova, Italy
Email:{rafiullah.khan, raffaele.bolla}@unige.it
Matteo Repetto, Roberto Bruschi
CNIT
Research Unit of University of Genova
16145 Genova, Italy
Email: {matteo.repetto, roberto.bruschi}@cnit.it
Maurizio Giribaldi
Infocom s.r.l,
P.zza Alessi 2/7
16128 Genova, Italy
Email: maurizio.giribaldi@infocomgenova.it
Abstract—Many network-based applications require full time
connectivity of the hosts for fate sharing and responding to
routine applications/protocols heart-beat messages. Past studies
revealed that network hosts are most of the time unused or idle
but still kept powered on just to maintain network connectivity.
Thus, significant energy savings are possible if the network
hosts can sleep when idle. Unfortunately, present low-power
sleep modes cannot maintain network connectivity and results
in applications state loss, thus preventing users from enabling
power management features. A proxy-based solution was recently
proposed in the literature that allows network hosts to sleep
and still maintain their network standings over Internet while
requiring minor changes to applications/protocols. The Network
Connectivity Proxy (NCP) can handle some basic network pres-
ence and management protocols like ICMP, DHCP, ARP etc on
behalf of sleeping hosts and wake them up only when it is truly
necessary. This paper addresses NCP generic architecture and
its requirements, main NCP responsibilities, different NCP types
and their characteristics and some possible solutions to preserve
open TCP connections for the sleeping hosts. It also describes key
challenges in the design and implementation of NCP and proposes
possible solutions. NCP can result in significant network energy
savings up to 70% depending on the hosts time usage model.
Index Terms—Green networking, energy efficiency, smart
proxy, power measurement.
I. INTRODUCTION
Green technology is one of the main challenges for the
today’s world to step into a sustainable society, that can
fulfill the needs of today and have the capability to meet
those of tomorrow. Main motivation for green technologies
is reducing the global CO2footprint [1]. Energy efficiency
in ICT became a high priority objective due to increase in
energy price, continuous growth of Internet-connected edge
devices, high data rates and increase in the number of services
offered by telecoms and Internet Service Providers (ISP).
Thus, green networking is a primary research topic for network
engineers to maintain the same performance and QoS of
current infrastructures while reducing energy consumption [2].
Recent research shows that energy consumption of network
end hosts is higher than servers and data centers and is
continuously growing due to the advancement in technology
and the increase in the number of network-based devices
like PCs, laptops, tablets, smart phones, game consoles, set-
top boxes etc. US Environmental Protection Agency (EPA)
estimated that network hosts consume about 2% of the overall
US electricity requirement [3].
Recent survey [4] revealed that about 60% to 70% of the
offices PCs are left powered on all the time even outside
work hours just to maintain network connectivity for remote
access, Instant Messaging (IM), Voice-over-IP (VoIP) clients,
file sharing, security updates or other administrative manage-
ment reasons. Most of network applications and protocols
require full-time network connectivity to receive/respond to
routine heart beat messages [5]. Present power management
sleep modes such as ACPI S3 and S4 states cannot maintain
network connectivity, thus network hosts are left powered on
to maintain full-time connectivity even when the hosts are idle,
which leads to significant wasted energy consumption [6].
This paper discusses the concept of Network Connectivity
Proxy (NCP) that has the capabilities to maintain full network
presence for sleeping hosts. The NCP should be always pow-
ered on and can be located on host’s Network Interface Card
(NIC) or other network entity that remains always powered
up such as routers, thus does not cause any increase in the
network energy consumption. NCP allows the network hosts
to sleep and still maintain their standing over Internet by
responding to applications/protocols heart-beat messages on
behalf of sleeping hosts. The basic idea of proxy was already
proposed in literature [2], [5], [7], [8]. This paper addresses
NCP generic architecture and requirements, characteristics of
different NCP types based on its functionality and location,
possible solutions to preserve open TCP connections for the
sleeping hosts and some other basic requirements and key
challenges in the design and implementation of the NCP.
The rest of the paper is organized as follows. Section II
describes the energy waste of ICT and current approaches
for low energy networking. Section III describes generic
proxy architecture, proxy requirements and types. Section
IV addresses some possible solutions to manage open TCP
connections. Section V addresses key challenges in the proxy
implementation. Section VI addresses expected energy savings
and finally, section VII concludes the paper and defines future
directions for proxying.
II. CURRENT APPROACHES TO REDUCE NETWORK
ENERGY WASTE
With the great advancement in the field of Information and
Communication Technology (ICT) and continuous growth of
Consumer Electronics (CE), the aggregate energy consumption
Fig. 1. Evoluation of router capacity vs. traffic volumes (Moore’s law) and
energy efficiency of silicon.
became a prime concern for the researchers. The silicon
technologies (e.g., CMOS) improve their energy efficiency by
a factor of 1.65 every 18 months as suggested by Dennard’s
scaling law. Fig. 1 shows that this improvement is low com-
pared to routers capacities (increase factor of 2.5/18 months)
and data traffic volumes increase which follows Moore’s law
and doubles after every 18 months [1]. The increasing energy
consumption has two negative impacts: economic and environ-
mental. Economic impact is obvious as the energy prices are
continuously increasing. Environmental impact comes from
the power plants that generate the electricity. These power
plants consume some precious natural resources and most of
them emit green house gases in the environment [1].
The International Energy Association (IEA) estimated en-
ergy consumption of ICT in Europe, North America, Pacific
and rest of the world as shown in Fig. 2. It is clear from
Fig. 2 that ICT and CE equipment energy consumption in
2012 is almost double compared to year 2005, and an increase
of almost three times is expected by year 2030 with the
current increase rate of energy requirement. The Global e-
Sustainability Initiative (GeSI) estimated an increase in carbon
footprint of ICT for year 2020 to about 320 Mtons of CO2
emission as shown in Fig. 3. The network infrastructure of
mobile communication and fixed wired narrow-band networks
cause most significant contributions to greenhouse gases.
Recent surveys revealed that most of the high capacity
routers and high data rate links are under-utilized most of
the time, thus causing considerable energy wastage. Several
ideas are proposed in the literature to reduce total network
energy consumption. These approaches can be classified into
four different categories.
A. Re-engineering
Re-engineering focuses on introducing and designing more
energy efficient elements for network devices. It consists of
new silicon (e.g., ASICs, FPGAs, network/packet processors,
etc) and memory technologies (Ternary Content-Addressable
Memory (TCAM), etc) for packet processing engines and
adoption of pure optical switching architecture [1]. Thus, re-
engineering focuses on optimization of internal device struc-
ture and reduction of complexity level.
Fig. 2. Estimated energy consumption of ICT and CE equipment.
Fig. 3. GeSI estimated emission of 𝐶𝑂2.
B. Hierarchy of hardware components
This approach introduces a hierarchy of power management
architectures for network hosts to reduce the energy consump-
tion [9]. It integrates several different power level computing
platforms into a single multi-tiered device. The device can
operate at only one tier or power level at a specific time. The
device uses low power tier for simpler tasks or under low load
conditions. It switches to higher power tiers if the processing
load increases. This approach significantly reduces energy
consumption and it can be very useful for mobile devices
that operate on battery power but at the cost of increasing
the device complexity.
C. Dynamic rate adaptation
Dynamic rate adaptation is based on varying network device
capabilities (e.g., link bandwidths, computational capacity of
packet processing engines, etc.) according to current traffic
loads and service requirements. Thus, dynamic rate adaptation
makes device energy efficient that consumes energy propor-
tional to its utilization [6]. It reduces working rate of the packet
processing engines by changing clock frequency and/or by
changing voltage of processors or by throttling the CPU clock.
Thus, it can be performed by using two power management
capabilities of the network devices called dynamic scaling
and idle logic. Dynamic scaling scales the voltage/frequency
to reduce or increase work rate according to current load
condition. Idle logic reduces power consumption by rapidly
turning off some sub-components in case of no activity and
re-waking them up when some new activities are received.
D. Smart Sleeping (Proxying)
Smart sleeping or proxying approach puts network devices
into sleep modes when they are idle and wakes them up
when it is truly necessary [10]. A special smart proxy can be
designed to maintain network presence for the sleeping hosts
[11]. Sleeping hosts consume much less power and higher
energy savings can be achieved.
III. NETWORK CONNECTIVITY PROXY
Many users leave their PCs powered on even outside
working hours when the PCs are largely idle, thus wasting
considerable amount of energy. There are two main reasons
behind this:
1) To maintain network connectivity for remote access, IM,
VoIP clients, security updates or other administrative
management reasons.
2) To share its resources (e.g. P2P file sharing).
A network connected host maintains its IP address and its
availability over the Internet. The proxy-based solution will
maintain network connectivity for the hosts when they are
going to sleep, thus it is important to understand the char-
acteristics of a fully connected network host [8]. A network
connected host should:
1) Respond to ARP requests in order to allow MAC level
communication in the network.
2) Maintain its IP address. In case of DHCP, it should
periodically generate DHCP lease requests in order to
maintain its IP address.
3) Respond to network diagnostic messages such as ping
that uses ICMP to detect the presence of host over the
network.
4) Maintain application-level reachability and allow new
TCP connections to be established by responding to TCP
SYN packets.
5) Maintain existing TCP connections.
6) Generate/respond to the network-based applications and
protocols periodic heart beat messages [7].
At present, ACPI standard sleep modes such as S3 (standby:
suspended to RAM) and S4 (hibernation: suspended to disk)
states are not able to maintain network connectivity. As a
result, these power saving features are not usually used by the
users. Power consumption of hosts can be reduced through
slow clock (dynamic rate adoptation) or stopped clock (sleep)
operation. Unfortunately, remote host access in stopped clock
mode is not possible as it is effectively sleeping [3]. The
NCP is designed for this purpose that implements key network
presence capabilities and maintains network standing for the
hosts when they enter into sleep mode. NCP starts functioning
on behalf of sleeping hosts and wake them up only when their
resources are required [5]. This is possible as modern NICs
have the capability to recognize magic packet and trigger host
wake up. The magic packet is used for the remote wake up of
sleeping host.
LAN or
Internet
Network Host Proxy
1
2
3
4
Fig. 4. NCP functional view: 1:host awake, 2:host going to sleep, 3:proxy
perform on behalf of host, 4:proxy wake up the host.
Fig. 4 shows a generic functional diagram for NCP. When
the host becomes idle and inactivity timer expires, it transfers
its network presence to the proxy and goes to sleep. The proxy
starts functioning on behalf of that host and receives/responds
to routine network traffic for the sleeping host. It wakes up the
host when a packet or TCP connection request is received that
requires host resources and returns the presence state back to
the host.
A. General Requirements
The NCP has some general requirements in order to allow
network hosts to sleep and save energy and still maintain their
standings over the network [8]. Some basic requirements of
NCP are as follows:
1) It should always remain fully powered up and present
in the network.
2) It should be in the same network/subnet as the host for
which it is covering. This will make easy for the NCP
to discover the power state of hosts and waking them
up using wake on LAN (WOL) packets.
3) It should be present on network devices which always
remain powered up such as routers/switches so that it
will not cause any significant increase in network energy
consumption.
4) It should be able to know about the power state of hosts.
5) It should be able to act on behalf of sleeping hosts by
using their IP addresses.
6) It should be able to handle simple network discovery
and management protocols such as ARP, ICMP, DHCP
etc on behalf of sleeping hosts.
7) It should be able to preserve existing TCP connections
for the sleeping hosts by generating and responding to
control messages.
8) It should be able to allow new TCP connection requests
for sleeping hosts.
9) It should be able to buffer incoming packets to give
enough time for the sleeping hosts to fully recover
from their stand-by modes. Buffers can also be used
to increase the host sleep time specially during the P2P
files download process. When the NCP buffer is almost
full, it wakes up the host, transfers data to the host and
then the host goes back again to sleep.
10) Generate/Respond to routine application/protocol mes-
sages as required by the application/protocol.
B. Modes of Operation
There are two different ways to operate the network con-
nectivity proxy [12].
1) Uncoordinated or Invisible Proxying: Uncoordinated or
invisible proxy doesn’t advertise its presence in the network. It
invisibly works in the network and guesses host’s power state
from the traffic analysis. When the host goes to sleep, proxy
invisibly starts functioning for it. The invisible proxy requires
host traffic to directly pass through it as it guesses host’s power
state from the traffic analysis. It does not require any change to
hosts and/or application servers. Since it doesn’t communicate
with the host, it cannot verify if the host permanently leave
the network while sleeping.
2) Coordinated or Cooperative Proxying: Coordinated or
cooperative proxy on the other hand announces its presence
in the network and communicates directly with the network
hosts. In cooperative proxying, NCP and host exchange two
kinds of messages: application specific and wakeup/sleep
messages. Application specific messages are used to register
host’s applications and services at the proxy. These contain
information about the application connections and application
routine messages. Wakeup/sleep messages are required to trig-
ger proxy when host goes to sleep and wake-up the host when
proxy receives a new connection request or a packet that needs
host resources. Thus, when a host goes to sleep based on power
saving policies or external user commands, it informs the
proxy about the power state change and transfers all proxyable
information like its MAC and IP address, port number of
all open sockets and connection sessions information, TCP
connection sequence and acknowledge numbers, etc. to the
proxy. The cooperative proxy can be located at any place in
the network and requires a software on the host that can be
used for communication with the proxy.
C. Types by Location
The proxying functionality can be placed in different de-
vices [13].
1) Self Proxy: In case of self proxying, NCP is located
within the network host on its NIC as shown in Fig. 5. It
can be invisible or cooperative proxy. NCP can use subset
of applications/protocols semantics with unique characteristics
that will generate responses without need of complex decision
process. NIC will proxy automated responses for network-
based applications and protocols and will allow host to sleep
when idle. This NCP has the advantage of sharing with the
host the same MAC and IP address. Modern NICs have an
auxiliary power source from the motherboard which allows
NICs to remain powered up even when the host is sleeping.
NIC consumes very low power, thus self proxying will not
cause any significant increase in network power consumption.
Computer
NIC
Proxy
Program
LAN or
Internet
Fig. 5. Self proxy.
Switch
Proxy
Program Internet
Network Host
Fig. 6. Switch proxy.
Since NIC has limited memory and computational power, thus
it will be difficult to implement a complete stand alone proxy
covering many applications.
Agarwal et al. [14] presented a USB based low power
architecture, named ‘Somniloquy’, that augments network
interfaces to enable sleeping PC to be responsive to network
traffic. Somniloquy can overcome the above limitations of self
proxying by embedding a secondary low power processor in
the PC’s network interface that runs an embedded operating
system. Somniloquy has significant memory and processing
capability and impersonates sleeping host by using same MAC
and IP address. Its flash storage is used to temporary store data
before being transferred in large chunks to the sleeping PC.
It has 64 MB DRAM and 2 GB flash that makes easier to
implement the proxying functionality.
2) Router/Switch Proxy: NCP can also be located on net-
work router/switch as shown in Fig. 6. Switch proxy can be
both, cooperative and invisible to the network. NCP will use
the WOL packet to wake up the sleeping host when its re-
sources are required. Since router/switch always remain pow-
ered up in the network, the NCP will not cause any significant
increase in the network power consumption. Better than host
NIC self proxying but still most of the routers/switches have
limited memory and computational power which will make
difficult to implement complete stand alone proxy covering
many applications for all of the network hosts.
3) Third Party Proxy: Third party proxy is a proxy located
on any other host in the network as shown in Fig. 4. This
NCP can only be cooperative and it requires communication
with the network hosts. It can have high memory and high
computational power but may increase the overall network
power consumption. Third party proxy can only be beneficial
if it is covering for very large number of network hosts.
Different approaches can be adopted to reduce network power
consumptions such as allowing NCP host itself to go to sleep
when idle or not covering for any host. Network hosts should
send first WOL packet and transfer proxyable state information
to NCP host before going to sleep.
4) Endpoint Proxy: The proxying functions are located on
remote endpoints of the host. When the host is going to sleep,
Switch
Proxy
Host Sleeping?
Packet Processor
Internet
Network Host
Action
Generator
Yes
No
Respond
Discard
Data or WOL
Buffer
Fig. 7. Generic proxy design.
it will inform the application endpoints about its power state
change with proxying request. The endpoints will apparently
disconnect application connections with the local host but will
preserve their state. When the host wakes up, it will contact the
endpoints and re-establish the proxied connections. Endpoint
proxy does not require any addition or change in the network
hardware but requires software modifications to both sides of
the connection. Endpoint proxy can solve the general NCP
problem of preserving TCP connections but every application
needs to have its own endpoint proxy. Irish and Christensen
in [15] proposed similar approach to preserve generic TCP
connections for the sleeping hosts.
5) Dual Proxy Approach: Dual proxy approach uses two
different proxies with different set of functionalities. It can be
used in the scenarios where single proxy does not have the
capability to meet all the requirements. Section V-D proposes
dual proxy approach to implement the proxying functionalities
over different networks/subnets.
D. General Design
The generic proxy block diagram is shown in Fig. 7. The
coordinated proxy has a socket connection to the network host
on which special control packets and power state information
flows. When the network host receives user command to go
to sleep, it transfers the power state change information and
other proxyable state information such as IP address, port
numbers of all open ports, sequence number of all open TCP
connections etc to the proxy. The invisible proxy requires the
proxy be on the path between the host and Internet. The
proxy simply directs packets to the network host when it
is awake otherwise it directs them to the packet processing
part [16]. When the host enters into sleep mode, the proxy
starts intercepting the packet directed to the sleeping host and
performs the appropriate action. Proxy generates simple ARP
and ICMP reply packets on behalf of sleeping host when
it receives the ARP or ICMP request packet. When proxy
receives packet directed to listening port but no appropriate
action is defined in proxy, it will sent WOL packet to wakeup
the sleeping host.
As shown in Fig. 7, the packet processing unit receives all
the packets directed to network host when it is sleeping. It
performs one of the following task depending on the type of
packet received.
1) Directly respond to packets such ARP, ICMP etc.
2) Wake up the host
Destination
MAC
6 Bytes
Source MAC
6 Bytes
Type
2 Bytes
6 bytes of 1, followed by dest. MAC
repeated 16 times
102 Bytes
Fig. 8. Generic format of WOL packet.
3) Discard the packet
4) Buffer the packets for later processing by the host when
it is awake.
The packet processing unit also generates the network
protocols/applications heart beat messages at periodic intervals
on behalf of sleeping host.
E. Wake on LAN
The wake on LAN (WOL) packet which is also known as
‘magic packet’ was invented by IBM in 1996 for the remote
waking up of a network connected PC through its NIC. All
modern NICs have an auxiliary power connection from the
motherboard. This allows NICs to remain powered up even
when the host is sleeping. NICs also have the connection to
the wake up interrupt line on the motherboard [3] used for the
host wake up. These NICs have the ability of pattern matching
for the received packets to trigger the host wake-up. The magic
packet is used for this purpose and it is sent in a MAC frame to
the sleeping host or broadcasted to all network hosts using the
network broadcast address. Thus, WOL packet wakes up the
host from ACPI S3 or S4 sleep states to fully powered on state
(S0). It contains in its payload six bytes of all 1’s followed by
the target computer’s MAC address (48-bit) repeated 16 times.
The generic format of the magic packet is shown in the Fig.
8. The following are some of the basic requirements of magic
packet:
1) It requires target computer MAC address.
2) It does not provide any delivery confirmation.
3) It requires target computer to be in the same sub-
net/network. WOL packet can also be sent over Internet
but requires to create new inbound firewall rule for
the router to accept packets directed to specific port
and broadcast them within the network. This increases
security risks such as denial of service attacks.
4) It requires target computer NIC to support pattern match-
ing and wake on LAN.
IV. MANAGING TCP CONNECTIONS
Many network-based applications such as telnet, ssh, and
other resources sharing applications establish permanent TCP
connections between client and server. TCP connection is
maintained even when no data is flowing by generating and
responding to periodic TCP keep alive messages. The main
challenging task for proxy is to get control of the host’s open
TCP connections when the host enters into sleep mode and
return the TCP connections back to the host when it wakes
up without any connection drop. The proxy should be able
to send TCP keep alive messages with the correct sequence
number according to the last received packet by the server. The
Actual
Server
Proxy
Server
Network
Host Internet
Fig. 9. Proxy server.
proxy should have the knowledge of sequence numbers and
acknowledgments for every open TCP connection of the host.
Below are some approaches that may be adopted to preserve
the TCP connections.
1) Re-writing Applications: The simplest way to preserve
the TCP connection is to re-write applications at both sides,
host and application server, so that they establish TCP connec-
tions only when they have data to transfer. Another possibility
can be allowing hosts to preserve the application state at the
application server when there is no active data transfer or
host is entering into sleep mode. This approach is usually not
appreciated as it requires changes to both hosts and application
server and it will also introduce some latency.
2) Proxy Servers: Proxy servers can be used as a new entity
at the application server side that embeds the capabilities of
preserving TCP connections/applications state as shown in
Fig. 9. Apparently, the proxy server is proxying the actual
application server instead of the network hosts. It can be
designed to perform a subset of tasks performed by the real
server. It intercepts all the requests directed to the application
server and check if it can fulfill the requests, otherwise forward
it directly to the actual server. The main motivation of proxy
server is to improve performance by reducing load on the real
server and having request filtering capability.
The proxy server maintains the TCP connection with the
real server on behalf of the sleeping hosts. When the host
wakes up, it contacts the proxy server and re-establish the
previous connection. The proxy server can be located close to
the server or it can act as part of the server. Every application
needs to have its own proxy server having the capability to
preserve the TCP connections with the actual server on behalf
of sleeping hosts.
3) Splitting TCP connections: The proxy splits TCP con-
nections between the host and the server as shown in Fig.
10. The host informs the proxy to preserve TCP connections
before going to sleep. During the sleep time, TCP connection
between the host and proxy is dropped but between proxy and
server will be still maintained. When the host wakes up due to
user activity or proxy WOL trigger, it will re-establish its TCP
connection with the proxy and will start using the existing TCP
connection with the server. Due to the limitations of self/switch
proxying, it will be difficult for proxy to manage many TCP
connections for different applications/hosts. Also this scheme
requires all TCP connections to be established through proxy
and it requires the proxy to be cooperative.
Christensen in [17] proposed the concept of split TCP
connection with the introduction of new ‘shim’ layer between
the application and socket interface. This approach uses the
existing TCP software socket layer and does not propose
any changes to application. This shim layer ‘fakes out’ the
TCP TCP
Host Prox
y
Server
Fig. 10. Split TCP connection.
application server to see the TCP connection established all
the time. When the host enters into sleep mode, the shim
layer at client informs the shim layer at server to drop
the TCP connection. When the host wakes up, the client
shim layer informs the server shim layer to re-establish the
connection with the socket information. This TCP splitting
requires additional software both at client and server but no
changes to the TCP.
4) Greening of TCP/IP: TCP connection can also be pre-
served for the sleeping host by introducing a new green
TCP/IP with a connection sleep option. This requires introduc-
ing a new connection sleep option in the TCP header. Sleep
option in client green TCP/IP will inform the server green
TCP/IP when the host is going to sleep. Server will keep the
connection open but it will not send any data/ACK to the
sleeping host [15]. Server also blocks the socket associated
with the TCP connection to prevent excessive data queue at
the server. When the host wakes up, it informs the server and
data flows can be resumed. The green TCP/IP is backward
compatible with normal TCP/IP as the normal TCP/IP ignores
any option it does not understand.
V. NCP KEY CHALLENGES
This section addresses some of the main limitations and key
challenges and proposes some possible solutions to be adopted
in the designing of proxy.
A. Memory and processing power requirements
This is the main limitation for self proxy and switch proxy.
Current NICs and switches don’t have enough memory and
processing power to implement complete standalone proxy
covering for many applications and network hosts. Third party
proxying or dual proxy approach can solve this problem but
it will increase the energy consumption as NCP is required to
be powered on all time. Third party proxying approach is only
useful if it is covering for large number of network hosts. The
power consumption can be reduced if the third party proxy
can also sleep during idle times and woken up by network
hosts using WOL packets before entering into sleep mode.
B. NCP location within network
The invisible proxy requires sleeping host to be directly
connected with NCP or traffic of the sleeping host should pass
through NCP. This limitation can only be addressed with the
help of cooperative proxy. The host informs the NCP about the
power state change along with proxyable information before
entering into the sleep mode. The NCP broadcasts the ARP
reply to the entire network and associate its MAC address with
the IP address of the proxied host. The NCP will broadcast
another ARP reply to re-associate proxied host MAC address
with host’s IP address when the host comes out of the sleep
state.
C. Host power state awareness
The NCP should be aware of the host presence and its
power state. Since the invisible proxy makes decisions from
the host’s traffic analysis, it is usually very difficult to know
about the host presence in the network after sleeping. The
NCP should be able to know if the host permanently leaves
the network and stops proxying for it. This problem may be
solved if the NCP wakes up the host at periodic intervals just
to check its presence in the network. However, this results
in un-necessary wake up and reduces the potential energy
savings. Modern NICs that support DMTF’s Alert Standard
Format (ASF) Specification 2.0 have the capability to respond
to ARP when the host is sleeping, thus making easy for the
proxy to check for host presence in the network.
D. NCP over differnt subnets/networks
The NCP can cover for network hosts that are located in
the same network/subnet. When the NCP receives a packet
that requires the sleeping host’s resources, it wakes up the
host using a WOL packet. The WOL packet contains the
target host MAC address in its payload. It is virtually im-
possible to have MAC-level communication between hosts in
different networks/subnets or to acquire the MAC addresses
of hosts from other subnets using ARP requests. Thus, the
host needs to communicate its MAC address to the proxy
before going to sleep. This problem can be addressed by using
the cooperative dual proxy approach and a concept similar
to ‘ARP proxy’ as shown in Fig. 11. The ARP proxy on
router allows the main proxy to communicate with the hosts of
other networks/subnets without being aware of the existence
of subnets. When host A wants to sent a packet to the main
proxy, it broadcasts an ARP request to know the MAC address
of the main proxy. The APR proxy on router will provide
ARP response on behalf of the main proxy with its own
MAC address. After the router receives a packet from the
host A, it forwards it to the main proxy. The router should
implement some additional proxying functionality that will
enable it to communicate directly with the main proxy and
get the information about all proxied sleeping hosts from it.
Whenever the router receives a packet for the sleeping host
A, it will forward it to the main proxy. In the same way,
the main proxy can wake up the sleeping host A using WOL
packet with the help of low functioning proxy at the router.
Since this scheme involves traffic forwarding, it may increase
the security risks.
E. Application independent proxy
This is one of the challenging tasks that need to be addressed
in the design of proxy. Every application has its specific
heartbeat/keep alive messages that need to be received and
responded at periodic intervals in order to keep the connection
open. The proxy needs to know the application and its session
data like ports, types of heartbeat messages, etc in order to
Internet
Router
Host A Main Proxy
ARP Proxy
Fig. 11. Proxying over different subnets.
maintain the application’s state for sleeping hosts. Due to
very large number of network-based applications and their
continuous increase, it would be practically impossible to
implement a proxy specific for each of them. The proxy should
be able to transparently discover what need to be proxied and
should require no modification to the applications.
F. NCP support for mobile hosts
The NCP should also be configured to allow host mobility
between networks. This possibility can be addressed with an
approach similar to ‘Mobile IP’ [18] and the proxy itself acts
as ‘Home Agent’ for the mobile host. Mobile IP allows host
to move from one network to another while maintaining its
permanent IP address. The mobile host is always identified
by its home address regardless of its current location in the
Internet. When the mobile host enters a new network, it is
assigned a care-of address which is associated to its home
address through a binding at the home agent. This scheme
will require the proxy to be strictly cooperative.
The NCP should also be able to support wireless mobile
hosts and handle with the connection disruption in wireless
environments which is the main design goal of invisible proxy.
Wireless mobile hosts may suffer from loss of connectivity
that results in dropping connections with the endpoint. The
proxy design should also take into consideration the wire-
less environment and propose solutions to handle short term
connection drops due to weak signal or any other reasons.
The cooperative NCP should periodically check if the mobile
host is alive, otherwise transparently start proxying for it for
predefined short time. If the host does not connect again to
the network within this short time, the proxy assumes that
host has permanently left the network and clears the proxied
resources.
VI. EXPECTED ENERGY SAVINGS
This section estimates possible energy saving by adopting
the NCP to allow network hosts to sleep during idle times.
The real energy savings estimation requires the knowledge
of all network-based devices, their active and sleep power
consumption and average use per day [19]. According to the
Survey presented in [4], almost about 60 to 70% people leave
their computers powered on at home and in the offices all
the time only to maintain network connectivity for resources
sharing, remote access and for other network centric appli-
cations. A normal desktop computer on the average requires
4 to 6W in the sleep mode and 60 to 80W of electricity to
be fully powered on. A full time powered on computer on
the average requires almost 650 kWh of electricity every year.
On the other hand, a computer powered on only for 8 hours
and sleep for the rest of the day requires almost 250 kWh of
electricity. The average cost of electricity in Europe is around
22 cent/kWh. Thus, a full time powered on PC consumes
around 145 Euro/year and NCP power managed PC consumes
around 55 Euro/year of electricity. Thus, a single computer on
the average can save around 90 Euro every year that accounts
to almost 60% savings. An organization with 10,000 PCs and
60% powered on all the time, will be able to save about 0.5
million Euro every year. Thus, enabling power management in
PCs can result in billion of Euro savings in the Europe alone
every year.
VII. CONCLUSIONS
This paper has discussed some possible network energy
saving approaches and particularly has focused on the network
connectivity proxy. The basic concept of NCP has been
described with reference to the previous work in literature.
Briefly, the generic structure of NCP and its main requirements
and responsibilities have been described. Also different types
of NCP along with benefits and limitations of each were
addressed. The existing solutions to preserve open TCP con-
nections for the sleeping hosts require modification to TCP/IP
stack. This paper addressed some possibilities to preserve TCP
connections. The main motivation of this paper is to present
a complete sketch of NCP to point out the key challenges in
its design and implementation and to propose some possible
solutions. NCP can result in significant network energy savings
about 60 to 70% depending on the network host’s time
usage model. Thus, NCP has the potential to provide great
environmental and economic savings in the field of ICT.
ACKNOWLEDGMENT
This work was partially supported by the European Com-
mission in the framework of the FP7 ECONET (low Energy
COnsumption NETworks) project (contract no. INFSO-ICT-
258454).
REFERENCES
[1] R. Bolla, R. Bruschi, F. Davoli, and F. Cucchietti, “Energy Efficiency
in the Future Internet: A Survey of Existing Approaches and Trends in
Energy-Aware Fixed Network Infrastructures,” in IEEE Communications
Surveys and Tutorials (COMST), vol 13 no. 2, pp. 223-244, May 2011.
[2] M. Gupta and S. Singh, “Greening of the Internet,” in Proceedings of
the 2003 conference on Applications, technologies, architectures, and
protocols for computer communications, SIGCOMM ’03, pp. 19-26,
New York, 2003.
[3] K. Christensen, P. Gunaratne, B. Nordman, and A. George, “The
Next Frontier for Communications Networks: Power Management,” in
Computer Communications, Vol. 27, No. 18, pp. 1758-1770, Dec. 2004.
[4] J. Roberson, C. Webber, M. McWhinney, R. Brown, M. Pinckard, and
J. Busch, “After-hours Power Status of Office Equipment and Energy use
of Miscellaneous Plugload Equipment,” in Lawrence Berkeley National
Laboratory, Berkeley, California. Report# LBNL-53729-Revised, 2004.
[5] K. Sabhanatarajan, A. Gordon-Ross, M. Oden, M. Navada, and
A. George, “Smart-NICs: Power Proxying for Reduced Power Con-
sumption in Network Edge Devices,” in IEEE Computer Society Annual
Symposium on VLSI, ISVLSI ’08, April 2008.
[6] S. Nedevschi, L. Popa, G. Iannaccone, S. Ratnasamy, and D. Wether-
all, “Reducing Network Energy Consumption via Sleeping and Rate-
Adaptation,” in Proceedings of the 5th USENIX Symposium on Net-
worked Systems Design and Implementation, NSDI ’08, USENIX Asso-
ciation Berkeley, CA, USA, 2008.
[7] M. Allman, K. Christensen, B. Nordman, and V. Paxson, “Enabling
an Energy-Efficient Future Internet Through Selectively Connected End
Systems,” in Sixth Workshop on Hot Topics in Networks (HotNets-VI),
November 2007.
[8] M. Jimeno, K. Christensen, and B. Nordman, “A Network Connection
Proxy to Enable Hosts to Sleep and Save Energy,” in IEEE International
Conference on Performance, Computing and Communications Confer-
ence (IPCCC), Dec. 2008.
[9] J. Sorber, N. Banerjee, M. D. Corner, and S. Rollins, “Turducken:
Hierarchical Power Management for Mobile Device,” in Proceedings of
the 3rd international conference on Mobile systems, applications, and
services, MobiSys ’05, 2005.
[10] S. Nedevschi, J. Chandrashekar, J. Liu, B. Nordman, S. Ratnasamy, and
N. Taft, “Skilled in the Art of Being Idle: Reducing Energy Waste in
Networked Systems,” in Proceedings of the 6th USENIX symposium
on Networked systems design and implementation, NSDI ’09, USENIX
Association Berkeley, CA, USA, 2009.
[11] K. Christensen and F. Gulledge, “Enabling Power Management for
Network-Attached Computers,” in Int. Journal of Network Management,
Vol. 8, No. 2, pp. 120-130, March-April 1998.
[12] J. Klamra, M. Olsson, K. Christensen, and B. Nordman, “Design and
Implementation of a Power Management Proxy for Universal Plug and
Play,” in Proceedings of the Swedish National Computer Networking
Workshop (SNCW 05), Sept. 2005.
[13] B. Nordman and K. Christensen, “Improving the Energy Efficiency
of Ethernet-Connected Devices: A Proposal for Proxying,” in Ethernet
Alliance White Paper, Sept. 2007.
[14] Y. Agarwal, S. Hodges, R. Chandra, J. Scott, P. Bahl, and R. Gupta,
“Somniloquy: Augmenting Network Interfaces to Reduce PC Energy
Usage,” in 6th ACM/USENIX Symp. On Networked Systems Design and
Implementation (NSDI 09), Boston, MA, USA, April 2009.
[15] L. Irish and K. Christensen, “A Green TCP/IP to Reduce Electricity
Consumed by Computers,” in Proceedings of IEEE Southeastcon, pp.
302-305, April 1998.
[16] P. Werstein and W. Vossen, “A Low-Power Proxy to Allow Unattended
Jabber Clients to Sleep,” in Ninth International Conference on Parallel
and Distributed Computing, Applications and Technologies, PDCAT ’08,
Dec. 2008.
[17] C. Gunaratne, K. Christensen, and B. Nordman, “Managing Energy
Consumption Costs in Desktop PCs and LAN Switches with Proxying,
Split TCP Connections, and Scaling of Link Speed,” in Int. Journal of
Network Management, Vol. 15, No. 5, pp. 297-310, September/October
2005.
[18] S. Cheshire and M. Baker, “Internet Mobility 4x4,” in Proceedings of
SIGCOMM, California, USA, August 1996.
[19] B. Nordman, M. A. Piette, and K. Kinne, “Measured Energy Savings
and Performance of Power-Managed Personal Computers and Monitors,”
in Proceedings of American Council for an Energy-Efficieny Economy
(ACEEE) summer study on energy efficiency in buildings, 1996.
