Design Requirements for Best Effort Backbone IP Network · Ethernet network, IP QoS design ... This...
-
Upload
truongdang -
Category
Documents
-
view
224 -
download
0
Transcript of Design Requirements for Best Effort Backbone IP Network · Ethernet network, IP QoS design ... This...
International Journal of Information and Electronics Engineering, Vol. 3, No. 3, May 2013
242DOI: 10.7763/IJIEE.2013.V3.308
Abstract—Process control systems for Hydrocarbon Process
Automation Applications (HPAA) are implemented in oil and gas fields, plants, refineries, and tank farms in the form of a Local Area Network (LAN) to support control functions. These systems are composed of sensors, actuators, and logic solvers networked together to form a control system platform. Reliable networking plays a key role in supporting such a system infrastructure. The existing network designs consist of multiple, parallel networks with limited interconnectivity supporting different functions. The concept of consolidating HPPA networks on a converged Internet Protocol (IP) and utilize Wide Area Network (WAN) for real time operation was not explored in detail in the past. This paper explores this concept by simulating a WAN network based on Best Effort Quality of Service settings. Simulation and empirical experimentation were conducted to assess the feasibility of such a conceptual design and it showed positive outcomes. HPPA can benefit from a converged IP WAN by minimizing network components and wiring; and provide an integrated control system platform at the end user's desktop.
Index Terms—Bandwidth, best effort (BE), converged IP, quality of service (QoS), peer-to-peer, traffic mix, WAN.
I. INTRODUCTION Process control systems for HPPA in a Wide Area
Network (WAN) is motivated by the increase in technology advancement in networking, which includes high speed Ethernet network, IP QoS design options, and the penetration of standard Ethernet interface into HPAA systems. Moreover, advancement in control systems logic solver, instrumentations, and the concept of distributed intelligence in HPAA systems drive the need for WAN network connectivity. HPAA applications can be in the form of control traffic in a Peer-to-Peer or multi-peer (s) to a master controller. Historically, these applications are based on dedicated and standalone networks with limited interconnection. Non-control applications such as voice, data, and media streaming; are typically supported by a separate infrastructure [1], [2].
The main characteristics of the HPAA LAN network are timeliness, availability, and reliability [1], [3]. These network attributes are essential to provide the foundation for supporting P2P control and safety systems. Timeliness in Process Control System (PCS) for an HPAA application can range from 10ms to 200sec. Hence the network shall be able to support the lower bound of the time delay requirements. Network availability is another key measure. The network
has to be close to 99.9999% for safety systems or 99.999% for other HPAA applications. Mix in availability requirements and the result is a network design that must be robust and sustainable at all times. Network reliability is essential in guaranteeing packet delivery and data integrity [1], [3]. Therefore, dedicated and standalone networks were designed and implemented over time to ensure both timeliness and a highly available network to support different process automation segments within an oil and gas plant.
Extensive work was completed on the timeliness of a real time dedicated network in the past. In addition, the network design robustness to ensure a highly available and reliable network was addressed by different network models, topologies and protocols. The concept of using a converged IP network for process control and non-control applications in a hydrocarbon operational Local Area Network environment was explored by the author [4]. Converged IP network for WAN has not yet been addressed. The performance and characteristics of such a network are not defined.
This paper is focused on exploring the BE IP WAN network for supporting both HPAA and non-HPAA applications. Typically, QoS can be based on BE and Priority based QoS settings. BE is where all the applications are allocated a bandwidth based on a weighted average of their traffic size. An application with large traffic demand will acquire more bandwidth than a lower one. Priority based QoS is where each application is assigned a unique priority indicator that governs how the application is treated by the network. Applications with higher priority will be processed faster than lower priority applications.
This paper is organized as follows: Background in section 2. Existing knowledge — by identifying researchers and their approach — in section 3. Discussion, simulation, results in section 4, Conclusion, with possible future work, is outlined in section 5.
II. BACKGROUND HPAA provides real-time performance for the
infrastructure supporting actual hydrocarbon process operation. This includes oil and gas wells, plants, pipelines network to ship the products to refineries, and tank farms. The overall process includes dealing with hydrocarbon products and their derivatives at high pressure, high temperature, and potentially dangerous processes and unprocessed petrochemicals [5].
By design, each local infrastructure work area (e.g., Process Instrumentation Buildings, shipper pumps yard, product distribution systems, etc.) are supported by a dedicated process control system segments to ensure a local
Design Requirements for Best Effort Backbone IP Network
Soloman M. Almadi, Ramzi El-Haddadeh, and Masaud Jahromi
Manuscript received August 15, 2012; revised October 19, 2012.Soloman M. Almadi is with Saudi Aramco (e-mail:
Ramzi El-Haddadeh is with Brunel Business School
Masaud Jahromi is with Ahlia University.
International Journal of Information and Electronics Engineering, Vol. 3, No. 3, May 2013
243
stfeprCprspTinefoau
Thinmcoimhaindicolocolaw
AWPlnenethapspchusneofvathtreEt
eady state opeed-back as anrocess segmenommunicationrocess contropan multiple imeliness andetwork. Convorm of a dedutonomous ge
III. EXIST
PCS networkhere are thre
nstrumentationmanagement inomplexity, mplementationave evolved nfrastructure iscussed inteommunicationowest layer oontrol, Fig. 1, ayers are now
with their pairs
Fig. The HPAA t
Area Network Wide Area Ne
laninsic (2004etworks and hetworking opthe characterizpplication perpeed, communharacteristics (sing a WAN networks that af Service deariance in netwhe heterogeneoends in estabthernet, are t
eration. Outpun input to ents in a form on network chal system. Thgeographic a
d reliability arentionally, thicated networographic area
TING RESEARCNET
ks for HPAA hee different n and device lnformation lev
expected ns. The instras a result and softwar
elligent noden enabled devof process auin the manufagenerating mo.
. 1. Process contr
traffic is most(LAN) and so
etwork (WAN4) [7], providehighlighted thtions. The papzation of thespective but hnication proto(fiber, wireles
network is WAare interconnesign criteria.
work reliabilitous nature of lishing homotransforming
ut from each pither downstr
of feedback coannels play a he communicareas in the e two key cha
his network isrk design and
a [4].
CH IN HPAA CTWORK have evolved onetworking
level to the covel. Each layperformance,rumentation aof developm
re applicationes (i.e., micvices) and theutomation inacturing field.ore traffic amo
rol layers [brooks
ly supported bome of this tr
N). Cucej, Gd a technology
he different loper did not prese different highlighted keocol methods, ss, etc.). The kAN may expanected without This cond
ty and timelinef the WAN ingenous netwowide area n
process segmream or upstontrol loops [5
key role in suation channelform of a Waracteristics o implementedd within a lim
CONVERGED IP
over the past ylayers conneontrol, and th
yer has its lev, and unand control l
ment in compns. Wilbankscroprocessor-beir adoption astrumentation. Hence, thesong each othe
s (2001)]
by dedicated Laffic is sent o
Gliecg, Kaisery brief of induocal and widerovide specifi
options fromey features suand their phy
key challenge nd over many
a uniform Qudition may ces especially d
nfrastructure. orks, such as network to b
ent is tream 5],[6]. uch a l can
WAN. of this d in a mited
P
years. ecting hen to vel of nique layers puting s [2] based at the n and e two
er and
Local over a r, and ustrial e area cs on m an uch as ysical when small uality create due to
New Giga
be an
optimTh
maxioperand geogothercommextenprocfieldconcand indunetwprocinforlocatconn
IV
Prconvdynainfrato apenvirmethbeenprimbehaperfoEffoand overtraffi
ThsimuWANand irepucustoswitcsuchcan bis usand dynajittercorre
InIP seincre80%undemeas
A.Fi
mal communiche benefits of imized when
rational plants control cent
graphically disr when neededmand and connsively and eess automatio
d) for differencept is not appgas) producin
ustries typicalwork with ded
ess. These nrmation resouted within tnected to each
V. CONVERGENETWORK BE
rocess controverged IP netwamics and dastructure anpproximate thronment. hod is well usn discussed c
mary objectiveavior utilizingormance agairt (BE). BE cease of suppo
rhead capacityfic load and pohe OPNET Nulation tool foN BE [13]. Oindustry and a
utation [13], [om applicatioches and trun
h media streambe utilized frosed to address
provide a hamics and netr, packet loss, elated to the tr
n this paper, thervices will bease the overa
% and then 10er these scenasure for both H
Simulationirst, the custom
cation platforma real-time Gused in a w
can be managter or differespersed; provid [9]. As a resntrol center is effectively us
on (i.e., withinnt plants aparplied in petrocng operating elly have a
dicated contronetworks are urce planningthe operating
h other [10].
ED IP NETWOREST EFFORT IP
ol system behwork is difficudependency od design [4],
he traffic behaSimulation a
sed in both acomprehensive
e of the simulg dedicated IPinst a converconverged IP ort [12]. The y is necessarossible traffic etwork Mode
or evaluating OPNET has exas a result has [14]. The too
ons (e.g., HPAnking plan). ming, IP telepm the OPNETcross networ
holistic perfotwork performand packet re
raffic load. he simulated Wbe exposed toall WAN trunk00%. The HParios providesHPAA and su
m build for H
m for PCS appiga-bit networ
wide area netged from a coment control ciding backup ssult, a distribu formed. Thissed in manag
n the factory orrt from each ochemical and henvironments.
standalone, ollers to mana
typically cog and managg facility an
RK CONVERGEP NETWORK Shavior for Hult to predict don the under[5], [6]. Simu
avior and netwas a research cademia and iely [11]. In lation is to evP network andrged IP netwwas selected fonly drawbac
ry to support surge [12].
eler package isthe concept o
xtensive use indeveloped its
ol enables thAA), network
Moreover, sphony, and laT application lrk traffic for mormance repomance parameetransmission
WAN networko an increase king utilizatio
PAA applicatis the anticipatupport services
HPAA applica
plications [8].rk can be furthtwork. Multipmmon commacenters that asupport for ea
uted autonomos concept can ging local plar plant operatiother [10]. Thhydrocarbon ( These differelocal real-tim
age a designatonnected to tgement systemnd are seldo
ED WIDE AREAIMULATION
HPAA in WAdue to the trafrlining netwoulation was uswork operation
and evaluatindustry and hthis paper, t
valuate the PCd benchmark
work based Bfor its simplicck is bandwidthe anticipat
s selected as tof converged n both academs s reliability ae user to bu
k elements (e.support servicarge file translibrary. The tomultiple servicorting. Trafeters such dela
are tracked a
k and convergtraffic load th
on from 50%,on performanted performans.
ation need to
. her ple and are ach ous be
ant ion his (oil ent me ted the ms om
A
AN ffic ork sed nal ion has the CS its est ity dth ted
the IP
mia and uild .g., ces fer ool ces ffic ay, and
ged hat to
nce nce
be
International Journal of Information and Electronics Engineering, Vol. 3, No. 3, May 2013
244
decocoreauesaninpepa
coU[1thRprth
A
m
ThdeCpatimsu
efined to refleontrollers. Thiontrollers andelationship). Tuthor as part stablished the nd define interncludes PCS ceriod and freqacket characte
The simulateonfirmed base
UDP/IP. TCP 15]. The HPAhe form of equest/Responrocessing timehe response for
B. SimulatioAssumptions In the simula
made for both W1) LAN ass
(2010) [2) WAN n
differentArea 2, PPCS Are
3) The back1 Gbps Elocation
4) HPAA Palong wiFTP.
5) Injected 80% and
Fig. 2. Proces
C. SimulatioThe PCS con
he focus is tedicated WANonverged IP nacket drop, Tmeout are vaubsequent sect
1) DedicatFig. 3 depict
ect its relationis includes thed controller tThe HPAA Pof Almadi, different elem
rnal and externcontroller procquency, and theristics. ed HPAA PCed on TCP/IP
provides betAA PCS contr
Request/Requnse model. The since the secr the first requ
on Network, C
ation model, tWide Area Nesumptions are4]. network simut HPAA PCSPCS Area 3 (Dea 4, and PCSkbone networEthernet linkswith BE QoS
PCS, voice anith the injected
traffic was ind 100% utiliza
ss control network
on Results ntroller applico establish a
AN network anetwork. DelayTCP retransmariable measutions. ted wide area s the applicati
nship betweene Master Contrto controller PCS controller
et al., (2010ments that gonal relationshcessing time, he nature of t
CS controller P transport prtter transmissiroller messagiuest/Responsehe later contricond request wuest has been
Controller and
the following etwork and Lo defined as pa
ulation is baS locations: PDefined as PC
S Area 5. rk, Fig. 2, is bas connecting e Converged IP
nd video servicd traffic for lo
creased systemation
k simulation mod
cation dynamiperformance
and compare y, jitter, retran
mission and Tres that will
network ion round trip
n the differentroller to subten(i.e., Peer-to
r is defined b0) [4]. The avern the contips. This definpacket size, cthe TCP and
was designedrotocol ratherion quality [1ing was define rather thanibutes to additwon’t processreceived.
d Application
assumptions ocal Area Netwart of Almadi,
ased defining PCS Area 1, CS Virtual Ce
ased on high seach of HPAAP network settces were simu
oading the netw
matically with
del for PCS mode
ics were simue benchmark
that to BE nsmission attemCP retransmibe utilized in
p delay betwee
t PCS nding
o-Peer by the author troller nition cyclic UDP
d and r than 14] & ned in n the tional s until
were work: et al.,
five PCS
enter),
speed A PCS
tings. ulated work,
h 50%,
el
lated. for a QoS
mpts, ission n the
en the
PCSArea
Fi
Fremomaxi
2Th
Contnodenode50%banda stePCSelemmedipairiPCSthe o
Thand A
VC and remas.
g. 3. PCSVC to P
ig. 4 depicts tote site contrimum delay is
Fig. 4. P
2) Convergedhe PCS Virttroller relatione and PCS Are, in between
%, 80%, and dwidth (1Gbpseady state for application
ments, intra-noia streaming ing. The PCS Area 2 is dep
order of secon
Fig. 5. P
he IP networkArea 2 is depic
mote site cont
PCS Areas Contro
the IP networrollers in thes below 33ms
PCSVC to PCS ar
d IP wide areatual Center (nships, PCS ea 2 that has
n. Simulated t100% netwo
s). The injecter 15 minutes
is communicode and inter-are establish
S application picted in Fig.5nds as a result
PCSVC to PCS ar
k delay betwected in Fig. 6.
trollers in the
oller Application
rk delay betwee different P.
reas IP network
a network (PCSVC) hArea 1 which at least one htraffic injectoork loading oed traffic load of simulationcating to the-node. The IPing their indeperformance
5. As noticed tof the 100% t
reas application
een PCSVC aThe maximum
e different PC
Response Delay
een PCSVC aCS Areas. T
delay
has two tighh is an adjacehop, a backbo
or was varied of the availabwas triggered
n time while te different PCP telephony aependent traffor PCSVC athe delay wastraffic load.
delay
and PCS Aream network del
CS
and The
htly ent one
at ble d at the CS and ffic and s in
a 1 lay,
atutThut
anapseou
apexexre80bu
nete80peww
t 100% utilizattilization at 26he 300ms netilization are c
Fig. 6. PC
Packet delivend 80% utilizapplication wasent and receivutcomes for pa
Fig. 7. P
Also, PCS apparent at the xtended over xpected, the etransmission 0% loading. Tuild-up is expe
Fi IP telephony
etwork utilizaelephony delay0% utilizatioerformance w
was in correlatiwas at 137ms.
0
10
20
30
40
50
60
70
seco
nd
s
tion, was 300m66ms. And, 33etwork delay considered hig
CSVC to PCS are
ery integrity wation since thes not the sameved are equalacket loss for
PCSVC to PCS ar
application TC80% and 100%40 seconds
100% loarate that start
This is an inderienced at hig
ig. 8. PCSVC to P
y encountered ation increasey was 765ms aon. At 50%
was also degraion to delay. AAt 80% it wa
MIN AVG
TCP Re
ms. This is fo3ms when the
at 100% angh and not acc
eas controller IP n
was apparentle packet sent ae. At 50% uti. Fig. 7 depithese three di
reas packet deliv
CP retransmi% utilization,
in both netwading resulteted five secondicator that ngh network lo
PCS areas TCP ti
performance s from 50%
at 100% as comthe delay
aded as utilizaAt 100%, the as at 119ms an
MAX
etransmission Timeout
50
80
10
llowed by theutilization at
nd 266ms at ceptable.
network delay
y suspected aand received foilization the picts the simulifferent scenar
ery integrity
ssion timeoutFig. 8 The timwork loadinged in an nds earlier thanetwork congeoading.
imeout
degradation ato 100%. Th
mpared to 730was 8ms.
ation increasedIP telephony
nd 0.028ms at
0% Network Utilization
0% Network Utilization
00% Network Utilization
e 80% 50%. 80%
at 100% or the
packet lation rios.
t was meout g. As early
an the estion
as the he IP
0ms at Jitter
d and jitter
t 50%
utilizutiliz
3Ne
applidelayfor apprrelatget mthe oPCSencodepeanalythe provnetw
Fig
Thfromdurinis aencoutilizfavoretrashutdthres
In perfoloadi
Benetw
zation. Table Ization.
TABLESupport Service Performance (mseconds) IP Telephony PacDelay IP Telephony PacDelay VariationCCTV Packet DeCCTV Jitter
3) Discussionetwork traffic lication and ny, packet lossBE is that e
ropriation baseted to its trafficmost of the baorder of 1024 application
ountered delayending on the ysis for BE codedicated IP
vided a closework. The 80%
g. 9. Dedicated vs
he TCP packetm zero at 50%ng a limited sian indicator ountered durization becamerable pattern
ansmission candown. Thereshold is essent
addition, suormance degring, depicted
est Effort Quwork is able to
I shows the av
E I: SUPPORT SER50% NetworLoad
cket 8
cket 0.028
elay 38 2.00E-0
n loading and mnetwork’s pers, and TCP reteach applicated on the weic volume. As aandwidth, as c4 bytes/second
— in a ys that span traffic load. F
onverged IP neP network. e network pe% and 100% sh
s. BE converged perform
t retransmissio% and 80% loa
imulation perioof the stru
ing data trae high. PCS and in fact
n be used as efore maintaitial for a seam
upport serviceradation at thein Table I.
V. CONC
uality of Serv support HPA
verage delay an
RVICES PERFORMA
rk 80% Network Load
730
119
1630 04 2.00E-03
mix have an imrformance vatransmission. tion will getighted averagea result, large compared to ad (i.e., PCS).converged Ifrom 0.28 mFig. 9 shows etwork at diffeThe 50% n
erformance tohow different
IP network applimance
on was increasad to over 38 od of 15 minuuggle the Pansmission, TCP retransmfor a safety a threshold t
ining utilizatmless PCS opees encounteree 80% and 10
CLUSION vice backbone
AA and other s
nd jitter at 100
ANCE 100% Network Load
765
137
1820 5.00E-02
mpact on the PCariables such
The conventit its bandwide that is direcfile transfer w
an application Therefore tP network
ms to 4 secondthe comparatierent loadings
network loadio the dedicatnetwork.
cation time delay
sed significancounts at 100
utes. This chanCS applicatiwhen netwo
mission is notsystem, pack
to trigger a sation at a loeration. ed unacceptab00% networki
e converged support servic
0%
CS as
ion dth ctly will n in the — ds, ive s to ing ted
y
ntly 0% nge ion ork t a ket afe ow
ble ing
IP es,
International Journal of Information and Electronics Engineering, Vol. 3, No. 3, May 2013
245
such as voice, media streaming, and large file transfer, if the network loading is maintained at 50% or below. The 50% overhead capacity provides the capacity for traffic surge. In addition this overhead capacity may be used during traffic rerouting caused by network outages.
The following are key guidelines to achieve the intended objectives:
1) PCS applications traffic load shall be estimated with at least 20% overhead growth factor.
2) IP telephony and media streaming traffic load shall be projected. Since these services are considered support services for industrial applications, their growth is not dynamic as compared to PCS HPAA application.
3) Media streaming operation is recommended to be on demand service rather than continuous streaming.
4) Design trunking plan to support both traffic surge and traffic reroutes where backbone trunks shall not exceed 50% network bandwidth utilization.
As part of future work, the priority based QoS shall be examined. Furthermore, the impacts of utilizing wireless backbone vs. wired shall be explored.
REFERENCES [1] H. Shah and S. Spada, “New Generation of Motion Control Networks
Fills the Gaps,” ARC Advisory Group, pp. 4-22, April 2005. [2] W. Wilbanks, “50 Years Of Progress In Measuring And Controlling
Industrial Processes,” IEEE Control Systems Magazine, vol. 16, no. 1, pp. 62, 64-66, Febrauray1996
[3] S. Boyer, SCADA: Supervisory Control and Data Acquisition, 3rd edn. ISA, 2004
[4] S. Almadi, R. E. Haddadeh, and J. Masaud, “Critical Requirements for Best Effort ConveregedIp Network For Hydrocarbon Process Automation Applications,” pp. 189-201, no. 7, 2010.
[5] R. Babatunde, A. Ogunnaike, and W. H. Ray, Process Dynamics, Modeling, and Control, Book, Oxford, 1994, 1994ISBN 0-19-509119-1
[6] Guideline for Safe Automation of Chemical Processes, AIChemE ISA 84.01 TR-002.
[7] Z. Cucej, D. Gleich, M. Kaiser, and P. Plannsic, “Industrial Networks,” in Proc. of 46th International Symposium Electronics in Marine, ELMAR, pp. 59-66, June 2004
[8] P. Brooks, “EtherNet/IP: Industrial Protocol,” in Proc. of 8th IEEE Emerging Technologies in Factory Automation, pp. 1-8, October 2001
[9] C. Brunner, “IEC 61850 Process Connection,” in Proc. of 15th Power Systems Computation Conference, pp.1-6, 2005.
[10] R. Cabezas, J. M. Selga, and C.Samitier, “A new Generation of Packet Switch Designed for the Integration of Operational Services,” in Proc. of CIGE Symposium, pp. 1-9, October 1999.
[11] R. Jain, The Art of Computer Systems Performance Analysis, New York: Wiley, 1991.
[12] J. Martin, “Managing Best Effort IP Networks over Heterogeneous WANs,” in Proc. of International Working Conference on
Performance Modeling and Evaluation of Heterogeneous Networks, pp. 1-6, July 2004
[13] J. Guo, W. Xiang, and S. Wang “Reinforce Networking Theory with OPNET Simulation,” Journal of Information Technology Education, pp. 215-226, 2007
[14] S. Almadi, R. E. Haddadeh, J. Masaud, “Converged IP Network For Hydrocarbon Process Automation,” in Proc. of IEEE Second International Conference on Computational Intelligence, Modelling and Simulation, pp. 1-6, September 2010
[15] S. Almadi, R. E. Haddadeh, F. Askandrani, and M. S. H. Jahromi, “Intelligent Field Converged IP Network for Semi-Real Hydrocarbon Process Automation Applications (HPAA) Case Study,” in Proc. of IEEE International Energy Conference, Manama, Bahrain, December 18-22, 2010
Soloman M. Almadi Soloman holds a PhD in computing from Brunel University, UK and Master of Science in Electrical Engineering from Southern Methodist University, US, andBachelor of Science in Electrical Engineering fromUniversity of Texas at Arlington, US. He is currently a professional Engineering Specialist in communication, process
control systems infrastructure, system integration, and Intelligent Field in Saudi Aramco, Saudi Arabia. He has over 20 years of experience at different capacities in the engineering, planning, and deployment of network and system solutions. Dr. Almadi is a member of IEEE, ISA, and has published several papers in networking, system and process automations, and simulations.
Ramzi El-Haddadeh is a faculty member at Brunel Business School. Prior to that; he was a faculty remember at The School of Information Systems Computing and Mathematics in Brunel University. He is a member of the Information Systems Evaluation and Integration Research Group (ISEing). He has published his work in peer reviewed and well acclaimed journals and international conferences. Dr. El-Haddahehcurrent
research interests include Networks operability telecommunication management, ICT adoption and diffusion and information security management in the public sector. He has guest-edited special issues of a number of international journals, and co-chairs a number of tracks at international conferences.
Masaud M. Jahromi was born in Manama-1970. BSC in Mechanical engineering from UOB-Bahrain 1994, MSC in Control and Information Technology from UMIST-Manchester 1996, PhD in Network Engineering from Kent University @ Canterbury-UK 2002. He worked as an Engineer in BAPCO-Bahrain for 6 months (1994). 1996-98
he used to work as a part-time engineer in A-A Computing Company-Manchester. 1998-2000 he used to work as a Lab administrator in Kent university @ Canterbury-UK. Since Feb. 2003 he is an instructor (Ass. Professor) in Ahlia university. Dr. Jahromi’sresearch are mainly focused in Networking, Network-Thinking, Queuing and traffic congestion, Network Security, and ad-hoc networks.
International Journal of Information and Electronics Engineering, Vol. 3, No. 3, May 2013
246