Research Article Minimizing Thermal Stress for Data Center ...downloads.hindawi.com/journals/tswj/2014/684501.pdf · Research Article Minimizing Thermal Stress for Data Center Servers

Research ArticleMinimizing Thermal Stress for Data Center Servers throughThermal-Aware Relocation

Muhammad Tayyab Chaudhry1 T C Ling1 S A Hussain2 and Atif Manzoor2

1 Universiti Malaya 50603 Kuala Lumpur Wilayah Persekutuan Malaysia2 COMSATS Institute of IT 54000 Lahore Punjab Pakistan

Correspondence should be addressed to Muhammad Tayyab Chaudhry mtayyabchyahoocomand T C Ling tchawumedumy

Received 20 December 2013 Accepted 12 February 2014 Published 31 March 2014

Academic Editor Chin-Chia Wu

Copyright copy 2014 Muhammad Tayyab Chaudhry et al This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

A rise in inlet air temperature may lower the rate of heat dissipation from air cooled computing servers This introduces a thermalstress to these servers As a result the poorly cooled active servers will start conducting heat to the neighboring servers andgiving rise to hotspot regions of thermal stress inside the data center As a result the physical hardware of these servers mayfail thus causing performance loss monetary loss and higher energy consumption for cooling mechanism In order to minimizethese situations this paper performs the profiling of inlet temperature sensitivity (ITS) and defines the optimum location for eachserver to minimize the chances of creating a thermal hotspot and thermal stress Based upon novel ITS analysis a thermal statemonitoring and server relocation algorithm for data centers is being proposed The contribution of this paper is bringing the peakoutlet temperatures of the relocated servers closer to average outlet temperature by over 5 times lowering the average peak outlettemperature by 35 and minimizing the thermal stress

1 Introduction

With the rapid proliferation of cloud services the data centerservers are experiencing increasing computational load eachyear The electrical power consumed by IT equipment isconverted into heat [1] An equal amount of power is requiredto remove that heat in order to maintain a proper workingenvironment via cooling mechanism The cooling mecha-nism works by blowing the cold air through hollow floor tilestowards server racks As a natural process the temperatureof cold air blown from the floor vents becomes more than theset temperature near the top of the racks In addition to thatthe hot air blown out from the air cooled servers from theback of the racks rises up and gets mixed with the cold airnear the top of the racks This recirculation of heat increasesthe cold air temperature as well [2ndash4]Thus the top mountedrack servers become the victims of inlet temperatureincrement

In a server which is a victim of high inlet temperaturethe heat removal efficiency is reduced Particularly when

servers are generating maximum heat at full utilizationthe hardware experiences thermal strain which changesto thermal stress [5] These poorly cooled servers startconducting heat to neighboring servers causing them tobecome undercooled Over a period of time the heatgenerated in the undercooled servers may exceed the rateof dissipation and a hotspot is formed Hotspots lead tohardware failure as well as performance loss and violation ofservice level agreement (SLA) In addition to this a hotspotdetected by data center thermal monitoring system maytrigger the cooling mechanism to cool down the hotspotthus leading to increased total cost of ownership [6] ofdata center

Heat dissipated by the servers depends upon their uti-lization levels and power consumption and can be markedby their outlet temperatures Heterogeneous servers dissipatedifferent amount of heat at same level of utilization andpower consumption This can be verified from the powerconsumption and heat dissipation statistics of the processorsas well The variation in inlet temperature has the typical

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 684501 9 pageshttpdxdoiorg1011552014684501

2 The Scientific World Journal

effect over heat dissipation of servers that can be profiled forinlet temperature sensitivity (ITS) This paper demonstratesthat the hotspots can be minimized if the servers are placedaccording to ITS profiling analysis Each server undergoesa thermal state transition on the basis of inlet temperaturevariations Hotspot is the extreme thermal state which laysthermal stress over servers The servers inside hotspots canbe relocated on the basis of same similar analysis to reducethe reoccurring of hotspots and to minimize thermal stress

This paper is organized as follows Section 2 shows therelated literature review Section 3 introduces the conceptsand terminology used in rest of the paper Section 4 describesthe ITS analysis with respect to thermal state transition andalso describes the thermal-aware relocation of data centerservers The experimental results and discussion are coveredby Section 5 Finally the conclusion is presented in Section 6

2 Related Work

Thermal modeling and temperature estimation [7 8] fromthermal sensors should consider that the increase in inlet airtemperaturemay cause some servers to undergo hotspot con-ditions and thermal stressThis is because they are not placedat proper positions according to thermal-aware locationanalysis Thermal-aware server provisioning approach withthe objective of minimizing the total power consumption ofdata center [4 9] calculates the power by considering themaximum working temperature of the servers Such calcula-tion should also consider that the inlet temperature rise maycause the servers to reach to the maximum temperature andcause thermal stress

Computational fluid dynamics (CFD) is popular tool forengineers Workload placement techniques that rely uponCFD simulations [2 4 10] can give the estimation of thermalstress besides the data center power consumption for coolingand computing if the location of servers the inlet tempera-ture variation and thermal-stress phenomenon are includedin the respective energy models A technique to reducerecirculation of hot air inside data center [2] can performbetter and save more cooling energy if the recirculation ofhot air is distinguished from the natural heating-up of coldair after it is blown from vent tiles However the factor ofreliability and thermal stress due to heat recirculation shouldalso be considered

The thermal data gathered from a range of thermalsensors will have the noise of acquired heat in cold air[11 12] due to physical phenomenon andor through heatrecirculation If this data is used for cooling control suchas implementing ASHRAE [13] standards then the serversshould be placed according to their thermal sensitivity toinlet temperature before data gathering could begin Since theASHRAE [13] requires the data center cooling temperatureto be increased doing so across data center will put someservers to go under thermal stress due to heat recirculationTherefore before making any decision regarding a raise incooling temperature the data center management shouldperform a thermal-stress evaluation for data center serversaccording to their location and inlet temperature

The coefficients of heat recirculation and heat extractionfor the data center servers [12] are sensitive to the inlettemperature increment and the value of coefficients shouldnot be affected by this phenomenon The data center work-load scheduling techniques by RC-thermal model of heatexchange [14 15] should consider that the backfilling of work-load may not work well if the change in inlet temperatureis not considered Additionally the backfilling can causehotspots and thermal stress upon the serves located in highinlet temperature region of data center Task-temperatureprofiles used for thermal-aware workload scheduling shouldconsider the effect of inlet temperature sensitivity of thephysical servers upon the scheduling outcome in terms ofthermal map to be unexpected [16]

The importance of arranging the servers according tothermal-stress analysis is that a thermal-aware workloadscheduling algorithm to consolidate active servers [17] willhave low chances of creating hotspots Similarly if the powerprofiles of servers are made as in [18] then they will have lesserrors if the profiling is performed after performing the serverarrangement for minimized thermal stress If the powersaving techniques such as diskless booting [19] are usedthen the servers will dissipate even less heat and undergoa minimum thermal stress if they are located in a thermal-aware arrangement

If the power consumption profiles of server are createdso that the least power is used to execute a given computingload and to ensure performance and profit as in [20] thenthe scheduling algorithm can savemore power if the hotspotsare avoided Additionally the monitory loss due to hardwarefailure can be avoided if the servers undergo minimumthermal stress The thermal profiling based techniques [2122] cannot give accurate results unless it is assured that theservers are efficiently placed across the data center in thermal-aware manner as proposed in this paper In order to achievea high thermostat setting for air conditioning [3 23] theproper placement of servers at optimum positions shouldbe prerequisite before evaluating the power consumption ofdata center Raising the cold air temperature can save coolingpower but it will increase thermal stress for the serversaffected by heat recirculation Eventually those servers at highutilization will experience thermal stress and may fail whilethe cooling mechanism might also be using more energy tocool down the hotspots

The data center power management and server con-solidation techniques can avoid hotspots thermal stressand unnecessary power usage for cooling by placing theservers at the most optimum positions according to thermal-stress analysis Scheduling algorithms to minimize heat canhave improved performance if the servers are placed atoptimum location to minimize thermal stress A server mayundergo various thermal states according to different factorssuch as thermal stress computational load and inlet airtemperature By identifying the thermal states of each serveran optimum location can be identified according to balanceof these factors and to minimum of the thermal stress Thispaper presents thermal state modeling approach for datacenter servers to identify and minimize the thermal stressthrough server relocation The benefits are reduced thermal

The Scientific World Journal 3

stress minimum hotspots and cooling energy saving in datacenters

3 Background

By the law of energy conservation the watts of electricalpower consumed are converted into equivalent joules ofthermal energy [1] If 119864119894computing is the electricity consumed bya data center server 119894 then this energy is converted to 119864119894Joules

119864119894

computing = 119864119894

Joules (1)

The air gets less cold when it travels towards servers afterbeing blown from the perforated tiles of hollow floor Thehike in inlet air temperature due to this and recirculation ofheat has a direct impact over outlet air temperature for eachserver So the outlet air temperature rises by the rise in inlettemperature But this relation is not strictly linear as the rise ininlet air temperaturemakes it a weaker conductor of heatTherate of heat transfer 119876119894 from server 119894 by conduction throughair [24] is given by

119876119894= 119896119894119860119894( 119879119894

server minus 119879119894

received) (2)

where 119896119894 is the coefficient of heat transfer for server 119894 119860119894 isthe overall area inside server through which the cold air attemperature 119879119894received flows and cools the server through con-duction and119879119894server is the overall temperature of the hardwareinside server casing The rise in inlet air temperature slowsdown the rate of heat transfer from the server dependingupon the make and model of the server The coefficient 119896119894may be different for heterogeneous servers The coefficient ofperformance (COP) curve [25] is unable to give a solution tothe situation when a server is getting hot due to rise in inletair temperature The server having high temperature of inletair119879received will have a corresponding increase in the outlet airtemperature as shown below

Δ119879119894= 119879119894

received minus 119879set (3)

where Δ119879119894 is the increase in inlet temperature of server 119894The highly dense arrangement of bare bone blade servers[26] can hold up to 96 servers in 45 u rack space Such adense existence of server can suffer fatal thermal stress whena server 119894 is exposed to increased inlet air temperature Δ119879119894 asshown in (3) The thermal stress 120590 [27] can be represented asfollows

120590119894= 1198641015840120572119894Δ119879119894 (4)

where 1198641015840 is the modulus of elasticity of the server 119894 and 120572119894 isthe coefficient of thermal expansion in mm∘C for server 119894The increase in inlet temperature causes thermal stress Overa period of time thismay eventually cause hardware failure asthe servers are tightly mounted in racksThe increased outlettemperature 119879119894outle(increased) of a server due to increase in inlettemperature has three effects

(i) First it puts extra burden on cooling mechanism asthe outlet temperature of the servers is increased

(ii) Secondly it may cause hotspot(iii) It may lay thermal-stress over server hardware

Data center servers have a built-inmechanism to dynamicallyadjust the outlet fan rotationwith respect to inlet temperatureas a reactive thermal management technique This is toincrease the airflow inside server casing tomaintain heat flowand to reduce the thermal stress The dynamic fan rotationcontrol may lead to loud noise andor hardware damage if thefan gets damaged [13] Some dynamic thermal management(DTM) routines apply frequency scaling to lower downprocessor speed in order to cool it down Disabling thedynamic fan control requires the inlet temperature to bewithin a vendor specified maximum value 119879119894max inlet

Consider a maximum threshold outlet temperature119879threshold from a server that is marked as hotspot temperatureby the monitoring system A subthreshold (119879119894max inlet minus 120573) isdefined for the indication of thermal stress The value for 120573 isnumeric and depends on COP

In highly dense arrangement of modern day bladeservers a rise in inlet air temperature followed by a risein computational load will put the servers in thermal-stressstate This will not only result in hotspots and equipmentfailure but also increase the data center wide cooling energyconsumption In particular at inlet temperature between(119879119894

max inlet minus 120573) and 119879119894

max inlet the server 119894 at high utiliza-tion will start experiencing thermal stress because DTMwill be inactive In this situation the outlet temperatureof a fully utilized server is maximized and may exceedthe peak temperature threshold 119879threshold At this time ahotspot is initiated by server 119894 It is important to lowerdown the peak outlet temperature of server 119894 to avoidhotspot To lower the peak outlet temperature it is bet-ter to relocate the server instead of shifting the work-load Server relocation provides a permanent solution tohotspots

4 Algorithm for Thermal-Aware ServerRelocation to Minimize Thermal Stress

This section presents the algorithm to reduce the thermalstress through thermal-aware server relocation those serverswhich arewere part of hotspot and those which are likelyto initiate hotspots are considered This paper is proposedto make a thermal profile of all the data center servers withrespect to inlet temperature Inlet temperature effect (ITE)thermal benchmark test can be used for this purpose ITEtest reveals the change in outlet temperature of a serverwith respect to changes in inlet temperature at zero and fullCPU utilization levels (In the rest of this paper the phraseserver utilization refers to CPU utilization because CPU isthe most power consuming and the most heat dissipatinghardware component of any computer system) The valuesof 119879119894max inlet and 119879119894outle(increased) can be inferred from ITEtest Homogenous servers have the same 119879119894max inlet However119879119894

outle(increased) depends upon the location of temperaturemonitoring sensors and can be verified with multiple testswith different sensor locations


Table 1 Domain table for thermal states

Thermal state 119879119894

inlet 119879119894

outlet 120583 120590119894 Chances of hotspot

119878119894

0119879119894

received asymp 119879set 119879119894

outlet lt 119879threshold asymp0 asymp0 Nil

119878119894

1119879119894


outlet lt 119879threshold gt0 asymp0 Nil

119878119894

2119879119894

received lt (119879119894

max inlet minus 120573) 119879119894

outlet le 119879threshold gt0 asymp0 Yes

119878119894

3((119879119894

max inlet minus 120573) minus 119879119894

received) le 0 119879119894

outlet ge 119879threshold gt0 gt0 Yes

4

3 26

15

7

8

910

1112

Si3

Si2

Si1

Si0

Figure 1 State transition diagram for data center server

41Thermal State Transition A finite set of thermal states fora server 119894 inside data center is demonstrated in the sectionwith respect to inlet temperature outlet temperature serverutilization level and thermal stress A thermal state of a server119894 can be defined as a tuple (119879119894inlet 119879

119894

outlet 120583 120590119894) and represented

by a notation 119878119894119899 where 119899 is a whole number and has a range

from 0 to 3 as per the state transition diagram demonstratedin Figure 1The domains of the elements of thermal states aredefined in Table 1 It is assumed that for all the states119879119894received lt119879thresholdThe states 119878119894

0and 1198781198941are the desired stateswhere there

is no thermal stress State 1198781198940is the idle state where the server

has no workload 120583 and the inlet temperature is close to theset temperature 119879set State 119878

119894

1is an active state of the server

where 120583 is not idle and the inlet temperature is same as thatof 1198781198940 Both these initial states have outlet temperature below

the red line temperature 119879thresholdThe difference between states 119878119894

2and 119878119894

3is that the

formal is an indicator of future thermal stress and futurehotspots State 119878119894

3may have thermal stress due to higher inlet

temperature compared to state 1198781198942 State 119878119894

3is the hotspot

state with outlet temperature being more than the maximumthreshold and the presence of thermal stress

Figure 1 demonstrates the state transition diagram whereall the states are at mesh The conditions for state transitionare given in Table 2 The desirable states are 119878119894

0to 1198781198941on the

basis of server utilization at Δ119879119894 equal to zero or minimumWhen Δ119879119894 becomes significant but remains lower thansubthreshold (119879119894max inlet minus 120573) at any state the yellow marked119878119894

2is reached This state is an indication of future thermal-

stress and likelihood of hotspot For any active server theviolation of (119879119894max inlet minus 120573) subthreshold represented by

((119879119894

max inletminus120573)minus119879119894

received) le 0 at any statemakes the respectiveserver reach to hotspot state 119878119894

3 An idle server with this

subthreshold violation is considered in state 1198781198940if the outlet

temperature is below 119879threshold By following the relocationalgorithmpresented in next subsection the servers from state119878119894

3can be brought to lower states and thermalstress can be

removedThis paper defines thermal profiling process consisting

of noting down the outlet temperature at minimum andmaximum utilization of server when the inlet temperatureis stable and below DTM threshold For each server 119894a thermal profile can be defined as a tuple having threeelements 119879119896max inlet 119905

119894

minCPU and 119905119894

maxCPU where the secondand third elements are equal to 119879119894outlet minus 119879

119894

received at minimumand maximum CPU utilization respectively The differencebetween 119905119894minCPU and 119905119894maxCPU shows the typical value ofmaximum increase in outlet temperature for any server when119879119894

received lt 119879119894

max inlet

42 Thermal-Aware Server Relocation Algorithm for Minimiz-ing Thermal Stress For each server at state 119878119894

3the relocation

algorithm can be given as in Algorithm 1The algorithm searches for a suitable server from the

set of server in state 1198781198940which can withstand the high inlet

temperature (listing (3)) Alternatively a server in state 1198781198941is

searched with more strict criteria that the maximum outlettemperature is below the hotspot server in addition to theinlet maximum temperature check (listing (8-9)) This isto make sure there will be no reoccurring hotspot afterswitching In case no server is found in lower states thehigher state 119878119894

2servers are searched with most strict criteria

that the minimum CPU outlet of 1198781198942server is lower than the

hotspot server (listing (14ndash16)) In case there is no suitableserver for location switching in the entire data center thealgorithm suggests shifting the workload from hotspot serverto a server in state 119878119894

0(listing (20)) Thus the proposed

algorithm can minimize the chances of hotspot of the serversin state 119878119894

3 The next section demonstrates the effectiveness of

location switching

5 Experimental Setup

Theproposed approachwas tested over a set of heterogeneousservers of make HP Paviolion ML350 G5 The servers haveVMware ESXi 50 [28] hypervisor installed Virtualizedservers (hosts) were used because the virtualization has


Table 2 Thermal states transition conditions table

Transition number Conditions1 120583 gt 0

2 Δ119879119894gt 0

3 ((119879119894


received) le 0 or (119879119894

outlet ge 119879threshold)

4 120583 asymp 0 | (Relocation and 120583 asymp 0)

5 120583 asymp 0

6 Δ119879119894asymp 0

7 Relocation and (Δ119879119894 asymp 0)

8 ((119879119894


received) le 0 and (120583 gt 0)

9 ((119879119894


received) le 0

10 (Δ119879119894asymp 0) and (120583 asymp 0)


12 (Δ119879119894gt 0) and (120583 gt 0)

Input(i) 1198783= All servers at state 119878119894

3

(ii) 1198780= All servers at state 119878119894

0

(iii) 1198781= All servers at state 119878119894

1

(iv) 1198782= All servers at state 119878119894

2

(v) 119878119894TTS= Thermal profiles and thermal states of all serversAlgorithm(1) For each server Server119894

1198783in set 119878

3

(2) Find a heterogeneous server Server1198951198780in set 119878119894

0for location switching such that

(3) 119878119895

TTS rarr 119879119895

max inlet ge 119878119894

TTS rarr 119879119894

received(4) If at least 1 server exists in 119878

0

(5) Switch locations of servers Server1198951198780With Server119894

1198783

(6) Else If no suitable server exists in 1198780



(8) 119878119896

TTS rarr 119879119896


TTS rarr 119879119894

received And(9) 119878

119896

TTS rarr 119905119896

maxCPU lt Server119894

1198783rarr 119905119894

maxCPU(10) If at least 1 suitable server exists in 119878

1

(11) Switch locations of servers Server1198941198781

With Server1198941198783




(14) 119878119898

TTS rarr 119879119898


TTS rarr 119879119894


119898

TTS rarr 119905119898


1198783rarr 119905119894

maxCPU

(16) 119878119898

TTS rarr 119905119898

minCPU lt Server119894

1198783rarr 119905119894

minCPU

(17) If at least 1 suitable server exists in 1198782


1198783


(20) Shift workload from Server1198941198783to server Server119895

1198780

(21) End

Algorithm 1


050100150200250300350

202530354045

1 12 23 34 45 56 67 78 89 100

111

122

133

144

155

166

177

188

199

210

221

232

243

254

Time (min)

ITE test for type A server

CPU usage ()Power (W)

CPU

usa

ge (

)po

wer

(W)

Inlet temperature (∘C)Outlet temperature (∘C)

Inle

t and

out

let

tem

pera

ture

(∘C)

Figure 2 ITE test performed for type A server

050100150200250300350

20

25

30

35

40

45

1 12 23 34 45 56 67 78 89 100

111

122

133

144

166

177

188

199

210

221

232

243

254

Time (min)

ITE test for type B server

CPU usage ()

Power (W)

Inlet temperature (∘C)

Outlet temperature (∘C)

Inle

t and

out

let

tem

pera

ture

(∘C)

CPU

usa

ge (

)po

wer

(W)

Figure 3 ITE test performed for type B server

a wide scope for cloud computing and virtualized datacenters The servers are classified as type A and type BType A has two quad core Intel(R) Xeon(R) CPU E5430266GHz processors Type B has Intel(R) Xeon(R) CPUE5320 186GHz processor Each server has 6GB of RAM Aset of 8 virtual machines (VMs) was executed on each serverduring the experiments All VMs were single virtual CPU(vCPU) withmaximum vCPU (limit kept in suspended state)with CPU intensive benchmark Prime95 [29] running overeach VMwhen suspendedThermal stress was introduced bythermostat settings of the air condition unit BY raising theset temperature a scenario of heat recirculation was createdduring the experiments

To monitor the inlet and outlet air temperatures externalUSB thermal sensors were used The power consumptionof each host was measured by USB smart power metersMicrosoft C script was used to manipulate the VM opera-tions such as powering on VMs with a specified batch sizeand VM suspension The servers are placed inside researchlab room with dimensions of 25 feet times 30 feet There are total10 desktops and 2 servers inside lab Two of the desktopsare Intel corei7 while the other 8 are Intel Pentium4 Eachdesktop has a standard size LCD There are two networkswitches and two wireless routers There are two split airconditioners inside lab with 2-ton cooling capacity each The

Table 3 Thermal variables gathered from ITE test

Server type Thermal variable Value (Celsius)A 119879

Amax inlet 26 to 27

B 119879Bmax inlet 25 to 26

A 120573 3B 120573 2A Average 119905AminCPU 11

B Average 119905BminCPU 13

020406080100120

2830323436384042

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)

SLI test for type A server

CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)

Figure 4 SLI test performed for thermal profiling of type A server

servers are placed under the table about 10 feet away from airconditionersThe tables are arranged horizontal to the airflowof air conditioners

51 Experimental Results and Discussion As a first stepITE test was performed by varying inlet temperature ofthe servers through thermostat setting at minimum andmaximum utilization of servers Figures 2 and 3 show theoutput of experiment The thermal variables gathered fromthe ITE tests are shown in Table 3

The maximum inlet temperature for both servers was setat 23 Celsius on the basis of results of ITE test as shownin Table 3 In order to profile the servers for 119905119894maxCPU thestep linear increment (SLI) test was performed In this testthe CPU intensive workload is put over servers in stepswhere each step involves the powering on of one VM aftera fixed interval of time such that the last step brings the CPUutilization of the host to maximum The inlet temperature iskept stable by placing the servers at a proper location Suchlocations were found by placing thermal sensors around theresearch lab and the readings were observed over few days tomark the suitable regions of room with stable temperature

Thermal profiles were created from two SLI tests atdifferent but stable inlet temperatures For first SLI test theaverage temperature was 213 Celsius which was well belowvalue of (119879119894max inletminus120573) subthresholdThe test results are shownin Figures 4 and 5

For the second SLI test the average inlet temperaturewas 235 Celsius which means that it is a hotspot causinginlet temperature given by (119879119894max inlet minus 120573) minus 119879

119894

received le 0


020406080100120

2830323436384042

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)

SLI test for type B server

CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)

Figure 5 SLI test performed for thermal profiling of type B server

Table 4 Thermal profile elements calculated from SLI tests

Server type Thermal variable Value (Celsius)A Average 119905AmaxCPU 18

B Average 119905BmaxCPU 19

A Average 119905AmaxCPU minus Average 119905AminCPU 7

B Average 119905BmaxCPU minus Average 119905BminCPU 6

Thermal profiles for the servers are shown in Table 4 whilethe test results are shown in Figures 6 and 7 If the inlettemperature remains stable and below the threshold of DTMa thermal profile for the servers can be created In thispaper the thermal profiles were created by SLI experimentsof Figures 4 and 5 and then verified later at hotspot causinginlet temperature in Figures 6 and 7 The results showthat the outlet temperature of a server can be predictedby extrapolation and interpolation of outlet temperatureswith respect to increase and decrease in inlet temperaturerespectively This paper verifies that the average peak outlettemperatures of the prototype servers can be extrapolatedwithin a range of average inlet temperature range 213ndash235Celsius The detailed results are available at [25]

511 Evaluation of State Transition Diagram By putting the119879threshold value to 42 Celsius and using the thermal profiles ofthe prototype servers the occurrence of hotspot and thermalstress can be verified Considering the servers in Figures 4and 5 were in states 119878119894

0and 1198781198941according to inlet temperature

then the servers were exposed to hotspot and thermal stresscausing inlet temperature in Figures 6 and 7 Comparisonof both sets of SLI test results shows that the type B servercan reach to state 119878119894

3as per thermal state transition diagram

of Figure 1 Type B server can be considered under thermalstress as average Δ119879119894 = 05 Celsius

Consider the server relocation algorithm by supposingthat type B server is active and has inlet temperature violatingsubthreshold (119879119894max inlet minus 120573) and the state of that server is119878119894

3 Also suppose that inlet temperature of type A server is

below subthreshold (119879119894max inletminus120573) and the server is in state 119878119894

1

Following the location switching algorithm the type B server

0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

Time (min)

SLI test II for type A server

CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)

Figure 6 SLI test for server type A for state determination andthermal profile verification

0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)

SLI test II for type B server

CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)

Figure 7 SLI test for server type B for state determination andthermal profile verification

can be relocated by location switching with type A server Soafter relocation both servers are at states lower than 119878119894

3 This

will also proves that the relocated servers will have

(i) homogenous outlet temperature despite differentinlet temperatures

(ii) no hotspot(iii) cooling cost saving

Plotting together the outlet temperatures of type A serverfrom Figure 4 and of type B server from Figure 7 intoFigure 8 shows that the outlet temperatures of servers arequite far from each other at all levels of utilization Focus onthe average peak outlet temperatures which are 35 Celsiusapart Before relocation as shown in Table 4 the outlettemperatures of both servers were almost the same plusmndifferentfrom the average peak of outlet temperatures If the serversare relocated the immediate effect is the homogeneity ofoutlet temperatures at all levels of utilization especially at thepeak and idle states By plotting the outlet temperatures ofboth servers form Figures 5 and 6 together in Figure 9 it canbe observed that the outlet temperatures of both servers are


Table 5 Average outlet temperatures of relocated servers before and after relocation

Beforeafterrelocation

Peak averagetemperature servertype A (Celsius)

Peak averagetemperature servertype B (Celsius)

Overall averagepeak temperature

(Celsius)

Difference from averagepeak temperature

(Celsius) server type A


(Celsius) server type BBefore 39 425 407 minus176 +174After 41 404 407 +023 minus035 Change 5 increase 5 decrease No change 765 improved 500 improved

3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)

Serversrsquo outlet temperature before relocation

Outlet temperature (∘C) server type AOutlet temperature (∘C) server type BAverage outlet temperature (∘C)

Out

let t

empe

ratu

re (∘

C)

Figure 8 The outlet temperatures from SLI tests of servers beforerelocation

closer to average temperature curve Thus the servers canbe relocated by using thermal profiles as one of the inputsparameter to relocation algorithm

Cooling cost is saved because all the relocated serversare not in state 119878119894

3to trigger the cooling system and the

peak average outlet temperature is reduced by 35 afterrelocation As demonstrated in Table 5 there is a significant5 to 765 times improvement in homogeneity of the averagepeak outlet temperatures of the servers after relocation Boththe servers are well below the maximum threshold of 42Celsius after relocation As a future work detailed thermalprofiles will be created and an outlet temperature predictiontechnique will be proposed on the basis of thermal profiles

6 Conclusion

This paper showed that the data center servers undergostate transition from normal to thermal stress as a resultof change in inlet temperature The servers can be profiledwith respect to ITS and outlet temperatures can be predictedfrom interpolation and extrapolation of thermal profilesThiswill be presented in more detail in future work The novelstate transition and ITS analysis for servers presented in thispaper manage to predict and track the state of a server whenthere is a change in inlet air temperature This is a novelpaper in which virtualized servers were used for thermal-stress and hotspot state evaluationThe servers inside hotspotarea can be relocated on the basis of ITS profiling and thermalstates based relocation algorithm presented in this paperThe algorithm can identify a more suitable location with

3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)

Serversrsquo outlet temperatures after relocation

Out

let t

empe

ratu

re (∘

C)Outlet temperature (∘C) server type AOutlet temperature (∘C) server type BAverage outlet temperature (∘C)

Figure 9 Outlet temperatures from SLI tests of the servers aremorehomogenous and close to average after relocation

minimum thermal stress for the hotspot affected serversThisrelocation process will avoid hotspots ensuring homogenousoutlet temperatures across the data center minimizing ther-mal stress lowering the peak average outlet temperatureand saving cooling power This paper shows that the peakaverage temperature was reduced by 35 and the peakoutlet temperatures of the relocated servers were closerto the average by over 5 times These results help in theestablishment of thermal-stress free green data centers

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgment

This research is supported by research Grant no RU018-2013from Universiti Malaya Malaysia

References

[1] GmbH LD Heat Heat as a form of energy Convertingelectrical energy into heat LD DIDACTIC GmbH Leyboldstr1 D-50354 Hurth Germany Germany p 1

[2] Q Tang S K S Gupta and G Varsamopoulos ldquoThermal-aware task scheduling for data centers through minimizingheat recirculationrdquo in Proceedings of the IEEE InternationalConference on Cluster Computing (CLUSTER rsquo07) pp 129ndash138


Austin Tex USA September 2007[3] A Banerjee T Mukherjee G Varsamopoulos and S K S

Gupta ldquoIntegrating cooling awareness with thermal awareworkload placement for HPC data centersrdquo Sustainable Com-puting vol 1 no 2 pp 134ndash150 2011

[4] T Mukherjee A Banerjee G Varsamopoulos S K S Guptaand S Rungta ldquoSpatio-temporal thermal-aware job schedulingto minimize energy consumption in virtualized heterogeneousdata centersrdquoComputer Networks vol 53 no 17 pp 2888ndash29042009

[5] J Wei ldquoThermal management of Fujitsursquos high-performanceserversrdquo Fujitsu Scientific and Technical Journal vol 43 no 1pp 122ndash129 2007

[6] Y Han ldquoCloud computing case studies and total costs ofownershiprdquo Information Technology and Libraries vol 30 no4 pp 198ndash206 2011

[7] M Jonas G Varsamopoulos and S K S Gupta ldquoNon-invasivethermal modeling techniques using ambient sensors for green-ing data centersrdquo in Proceedings of the 39th InternationalConference on Parallel Processing Workshops (ICPPW rsquo10) pp453ndash460 San Diego Calif USA September 2010

[8] M Jonas GVarsamopoulos and S K S Gupta ldquoOndevelopinga fast cost-effective and non-invasive method to derive datacenter thermal mapsrdquo in Proceedings of the IEEE InternationalConference on Cluster Computing (CLUSTER rsquo07) pp 474ndash475Austin Tex USA September 2007

[9] Z Abbasi G Varsamopoulos and S K S Gupta ldquoThermalaware server provisioning and workload distribution for inter-net data centersrdquo in Proceedings of the 19th ACM Interna-tional Symposium on High Performance Distributed Computing(HPDC rsquo10) pp 130ndash141 Chicago Ill USA June 2010

[10] N Ahuja ldquoDatacenter power savings through high ambientdatacenter operation CFD modeling studyrdquo in Proceedings ofthe 28thAnnual IEEE SemiconductorThermalMeasurement andManagement Symposium (SEMI-THERM rsquo12) San Jose CalifUSA March 2012

[11] N Ahuja C Rego S Ahuja M Warner and A DoccaldquoData center efficiency with higher ambient temperatures andoptimized cooling controlrdquo in Proceedings of the 27th AnnualIEEE Semiconductor Thermal Measurement and Management(SEMI-THERM rsquo11) pp 105ndash109 San Jose Calif USA March2011

[12] Q Tang T Mukherjee S K S Gupta and P Cayton ldquoSensor-based fast thermal evaluation model for energy efficient high-performance datacentersrdquo inProceedings of the 4th InternationalConference on Intelligent Sensing and Information Processing(ICISIP rsquo06) pp 203ndash208 Bangalore India December 2006

[13] ASHRAE-TC-9 9 2011 Thermal Guidelines for Data Process-ing EnvironmentsmdashExpanded Data Center Classes and UsageGuidance 2011

[14] L Wang G von Laszewski J Dayal X He A J Younge and TR Furlani ldquoTowards thermal aware workload scheduling in adata centerrdquo in Proceedings of the 10th International Symposiumon Pervasive Systems Algorithms andNetworks (ISPAN rsquo09) pp116ndash122 Kaohsiung Taiwan December 2009

[15] LWangGVonLaszewski J Dayal andT R Furlani ldquoThermalaware workload scheduling with backfilling for green data cen-tersrdquo in Proceedings of the IEEE 28th International PerformanceComputing and Communications Conference (IPCCC rsquo09) pp289ndash296 Phoenix Ariz USA December 2009

[16] L Wang S U Khan and J Dayal ldquoThermal aware workloadplacement with task-temperature profiles in a data centerrdquoTheJournal of Supercomputing vol 61 no 3 pp 780ndash803 2012

[17] A CorradiM Fanelli and L Foschini ldquoIncreasing cloud powerefficiency through consolidation techniquesrdquo in Proceedings ofthe 16th IEEE Symposium on Computers and Communications(ISCC rsquo11) pp 129ndash134 Kerkyra Greece July 2011

[18] Z Zhang and S Fu ldquoProfiling and analysis of power consump-tion for virtualized systems and applicationsrdquo in Proceedingsof the IEEE 29th International Performance Computing andCommunications Conference (IPCCC rsquo10) pp 329ndash330 Albu-querque NM USA December 2010

[19] C Y Tu W C Kuo W H Teng Y T Wang and S ShiauldquoA power-aware cloud architecture with smart meteringrdquo inProceedings of the 39th International Conference on ParallelProcessing Workshops (ICPPW rsquo10) pp 497ndash503 San DiegoCalif USA September 2010

[20] D Kusic J O Kephart J E Hanson N Kandasamy andG Jiang ldquoPower and performance management of virtualizedcomputing environments via lookahead controlrdquo Cluster Com-puting vol 12 no 1 pp 1ndash15 2009

[21] I Rodero H Viswanathan E K Lee et al ldquoEnergy-efficientthermal-aware autonomic management of virtualized HPCcloud infrastructurerdquo Journal of Grid Computing vol 10 no 3pp 447ndash473 2012

[22] I Rodero E K Lee D Pompili M Parashar M Gamell andR J Figueiredo ldquoTowards energy-efficient reactive thermalmanagement in instrumented datacentersrdquo in Proceedings of the11th IEEEACM International Conference on Grid Computingpp 321ndash328 Brussels Belgium October 2010

[23] A Banerjee T Mukherjee G Varsamopoulos and S K SGupta ldquoCooling-aware and thermal-awareworkload placementfor green HPC data centersrdquo in Proceedings of the InternationalConference onGreenComputing pp 245ndash256 Chicago Ill USAAugust 2010

[24] T Henderson Heat and Temperature Rates of Heat Transfer2013 httpwwwphysicsclassroomcomclassthermalPu18l1fcfm

[25] ldquoDetailed test results password lsquostressrsquordquo httpswwwmediafirecomfolderkks5249iaoimpThermal-stress

[26] S Stanley Roving Reporter Driving Up Server Density2012 httpembeddedcommunitiesintelcomcommunityenhardwareblog20120913roving-reporter-driving-up-server-density

[27] R Verterra ldquoThermal-stressrdquo httpwwwmathalinocomrevi-ewermechanics-and-strength-of-materialsthermal-stress

[28] VMware VMware vSphere Basics ESXi 5 0 vCenter Server 5 02009

[29] Great Internet Mersenne Prime Search (GIMPS) 2012httpwwwmersenneorgfreesoft

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



effect over heat dissipation of servers that can be profiled forinlet temperature sensitivity (ITS) This paper demonstratesthat the hotspots can be minimized if the servers are placedaccording to ITS profiling analysis Each server undergoesa thermal state transition on the basis of inlet temperaturevariations Hotspot is the extreme thermal state which laysthermal stress over servers The servers inside hotspots canbe relocated on the basis of same similar analysis to reducethe reoccurring of hotspots and to minimize thermal stress

This paper is organized as follows Section 2 shows therelated literature review Section 3 introduces the conceptsand terminology used in rest of the paper Section 4 describesthe ITS analysis with respect to thermal state transition andalso describes the thermal-aware relocation of data centerservers The experimental results and discussion are coveredby Section 5 Finally the conclusion is presented in Section 6

2 Related Work

Thermal modeling and temperature estimation [7 8] fromthermal sensors should consider that the increase in inlet airtemperaturemay cause some servers to undergo hotspot con-ditions and thermal stressThis is because they are not placedat proper positions according to thermal-aware locationanalysis Thermal-aware server provisioning approach withthe objective of minimizing the total power consumption ofdata center [4 9] calculates the power by considering themaximum working temperature of the servers Such calcula-tion should also consider that the inlet temperature rise maycause the servers to reach to the maximum temperature andcause thermal stress

Computational fluid dynamics (CFD) is popular tool forengineers Workload placement techniques that rely uponCFD simulations [2 4 10] can give the estimation of thermalstress besides the data center power consumption for coolingand computing if the location of servers the inlet tempera-ture variation and thermal-stress phenomenon are includedin the respective energy models A technique to reducerecirculation of hot air inside data center [2] can performbetter and save more cooling energy if the recirculation ofhot air is distinguished from the natural heating-up of coldair after it is blown from vent tiles However the factor ofreliability and thermal stress due to heat recirculation shouldalso be considered

The thermal data gathered from a range of thermalsensors will have the noise of acquired heat in cold air[11 12] due to physical phenomenon andor through heatrecirculation If this data is used for cooling control suchas implementing ASHRAE [13] standards then the serversshould be placed according to their thermal sensitivity toinlet temperature before data gathering could begin Since theASHRAE [13] requires the data center cooling temperatureto be increased doing so across data center will put someservers to go under thermal stress due to heat recirculationTherefore before making any decision regarding a raise incooling temperature the data center management shouldperform a thermal-stress evaluation for data center serversaccording to their location and inlet temperature

The coefficients of heat recirculation and heat extractionfor the data center servers [12] are sensitive to the inlettemperature increment and the value of coefficients shouldnot be affected by this phenomenon The data center work-load scheduling techniques by RC-thermal model of heatexchange [14 15] should consider that the backfilling of work-load may not work well if the change in inlet temperatureis not considered Additionally the backfilling can causehotspots and thermal stress upon the serves located in highinlet temperature region of data center Task-temperatureprofiles used for thermal-aware workload scheduling shouldconsider the effect of inlet temperature sensitivity of thephysical servers upon the scheduling outcome in terms ofthermal map to be unexpected [16]

The importance of arranging the servers according tothermal-stress analysis is that a thermal-aware workloadscheduling algorithm to consolidate active servers [17] willhave low chances of creating hotspots Similarly if the powerprofiles of servers are made as in [18] then they will have lesserrors if the profiling is performed after performing the serverarrangement for minimized thermal stress If the powersaving techniques such as diskless booting [19] are usedthen the servers will dissipate even less heat and undergoa minimum thermal stress if they are located in a thermal-aware arrangement

If the power consumption profiles of server are createdso that the least power is used to execute a given computingload and to ensure performance and profit as in [20] thenthe scheduling algorithm can savemore power if the hotspotsare avoided Additionally the monitory loss due to hardwarefailure can be avoided if the servers undergo minimumthermal stress The thermal profiling based techniques [2122] cannot give accurate results unless it is assured that theservers are efficiently placed across the data center in thermal-aware manner as proposed in this paper In order to achievea high thermostat setting for air conditioning [3 23] theproper placement of servers at optimum positions shouldbe prerequisite before evaluating the power consumption ofdata center Raising the cold air temperature can save coolingpower but it will increase thermal stress for the serversaffected by heat recirculation Eventually those servers at highutilization will experience thermal stress and may fail whilethe cooling mechanism might also be using more energy tocool down the hotspots

The data center power management and server con-solidation techniques can avoid hotspots thermal stressand unnecessary power usage for cooling by placing theservers at the most optimum positions according to thermal-stress analysis Scheduling algorithms to minimize heat canhave improved performance if the servers are placed atoptimum location to minimize thermal stress A server mayundergo various thermal states according to different factorssuch as thermal stress computational load and inlet airtemperature By identifying the thermal states of each serveran optimum location can be identified according to balanceof these factors and to minimum of the thermal stress Thispaper presents thermal state modeling approach for datacenter servers to identify and minimize the thermal stressthrough server relocation The benefits are reduced thermal



3 Background


119864119894

computing = 119864119894

Joules (1)


119876119894= 119896119894119860119894( 119879119894


received) (2)


Δ119879119894= 119879119894



120590119894= 1198641015840120572119894Δ119879119894 (4)















inlet 119879119894


119878119894

0119879119894



119878119894

1119879119894



119878119894

2119879119894


max inlet minus 120573) 119879119894


119878119894

3((119879119894


received) le 0 119879119894


4

3 26

15

7

8

910

1112

Si3

Si2

Si1

Si0



119894







119894




2and 119878119894

3is that the




3is the hotspot



0to 1198781198941on the




((119879119894









119894

minCPU and 119905119894


119894



max inlet


3the relocation











location switching






2 Δ119879119894gt 0

3 ((119879119894


received) le 0 or (119879119894



5 120583 asymp 0

6 Δ119879119894asymp 0


8 ((119879119894



9 ((119879119894


received) le 0

10 (Δ119879119894asymp 0) and (120583 asymp 0)


12 (Δ119879119894gt 0) and (120583 gt 0)


3


0


1


2


1198783in set 119878

3



(3) 119878119895

TTS rarr 119879119895


TTS rarr 119879119894


0


1198783




(8) 119878119896

TTS rarr 119879119896


TTS rarr 119879119894


119896

TTS rarr 119905119896


1198783rarr 119905119894


1






(14) 119878119898

TTS rarr 119879119898


TTS rarr 119879119894


119898

TTS rarr 119905119898


1198783rarr 119905119894

maxCPU

(16) 119878119898

TTS rarr 119905119898


1198783rarr 119905119894

minCPU



1198783



1198780

(21) End

Algorithm 1


050100150200250300350

202530354045

1 12 23 34 45 56 67 78 89 100

111

122

133

144

155

166

177

188

199

210

221

232

243

254

Time (min)



CPU

usa

ge (

)po

wer

(W)


Inle

t and

out

let

tem

pera

ture

(∘C)


050100150200250300350

20

25

30

35

40

45

1 12 23 34 45 56 67 78 89 100

111

122

133

144

166

177

188

199

210

221

232

243

254

Time (min)


CPU usage ()

Power (W)



Inle

t and

out

let

tem

pera

ture

(∘C)

CPU

usa

ge (

)po

wer

(W)






Amax inlet 26 to 27




020406080100120

2830323436384042

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)







119894

received le 0


020406080100120

2830323436384042

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)
















1


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)



3 This











(Celsius)





3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)



Out

let t

empe

ratu

re (∘

C)






6 Conclusion


3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


Out

let t

empe

ratu

re (∘






Acknowledgment


References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





3 Background


119864119894

computing = 119864119894

Joules (1)


119876119894= 119896119894119860119894( 119879119894


received) (2)


Δ119879119894= 119879119894



120590119894= 1198641015840120572119894Δ119879119894 (4)















inlet 119879119894


119878119894

0119879119894



119878119894

1119879119894



119878119894

2119879119894


max inlet minus 120573) 119879119894


119878119894

3((119879119894


received) le 0 119879119894


4

3 26

15

7

8

910

1112

Si3

Si2

Si1

Si0



119894







119894




2and 119878119894

3is that the




3is the hotspot



0to 1198781198941on the




((119879119894









119894

minCPU and 119905119894


119894



max inlet


3the relocation











location switching






2 Δ119879119894gt 0

3 ((119879119894


received) le 0 or (119879119894



5 120583 asymp 0

6 Δ119879119894asymp 0


8 ((119879119894



9 ((119879119894


received) le 0

10 (Δ119879119894asymp 0) and (120583 asymp 0)


12 (Δ119879119894gt 0) and (120583 gt 0)


3


0


1


2


1198783in set 119878

3



(3) 119878119895

TTS rarr 119879119895


TTS rarr 119879119894


0


1198783




(8) 119878119896

TTS rarr 119879119896


TTS rarr 119879119894


119896

TTS rarr 119905119896


1198783rarr 119905119894


1






(14) 119878119898

TTS rarr 119879119898


TTS rarr 119879119894


119898

TTS rarr 119905119898


1198783rarr 119905119894

maxCPU

(16) 119878119898

TTS rarr 119905119898


1198783rarr 119905119894

minCPU



1198783



1198780

(21) End

Algorithm 1


050100150200250300350

202530354045

1 12 23 34 45 56 67 78 89 100

111

122

133

144

155

166

177

188

199

210

221

232

243

254

Time (min)



CPU

usa

ge (

)po

wer

(W)


Inle

t and

out

let

tem

pera

ture

(∘C)


050100150200250300350

20

25

30

35

40

45

1 12 23 34 45 56 67 78 89 100

111

122

133

144

166

177

188

199

210

221

232

243

254

Time (min)


CPU usage ()

Power (W)



Inle

t and

out

let

tem

pera

ture

(∘C)

CPU

usa

ge (

)po

wer

(W)






Amax inlet 26 to 27




020406080100120

2830323436384042

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)







119894

received le 0


020406080100120

2830323436384042

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)
















1


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)



3 This











(Celsius)





3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)



Out

let t

empe

ratu

re (∘

C)






6 Conclusion


3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


Out

let t

empe

ratu

re (∘






Acknowledgment


References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






inlet 119879119894


119878119894

0119879119894



119878119894

1119879119894



119878119894

2119879119894


max inlet minus 120573) 119879119894


119878119894

3((119879119894


received) le 0 119879119894


4

3 26

15

7

8

910

1112

Si3

Si2

Si1

Si0



119894







119894




2and 119878119894

3is that the




3is the hotspot



0to 1198781198941on the




((119879119894









119894

minCPU and 119905119894


119894



max inlet


3the relocation











location switching






2 Δ119879119894gt 0

3 ((119879119894


received) le 0 or (119879119894



5 120583 asymp 0

6 Δ119879119894asymp 0


8 ((119879119894



9 ((119879119894


received) le 0

10 (Δ119879119894asymp 0) and (120583 asymp 0)


12 (Δ119879119894gt 0) and (120583 gt 0)


3


0


1


2


1198783in set 119878

3



(3) 119878119895

TTS rarr 119879119895


TTS rarr 119879119894


0


1198783




(8) 119878119896

TTS rarr 119879119896


TTS rarr 119879119894


119896

TTS rarr 119905119896


1198783rarr 119905119894


1






(14) 119878119898

TTS rarr 119879119898


TTS rarr 119879119894


119898

TTS rarr 119905119898


1198783rarr 119905119894

maxCPU

(16) 119878119898

TTS rarr 119905119898


1198783rarr 119905119894

minCPU



1198783



1198780

(21) End

Algorithm 1


050100150200250300350

202530354045

1 12 23 34 45 56 67 78 89 100

111

122

133

144

155

166

177

188

199

210

221

232

243

254

Time (min)



CPU

usa

ge (

)po

wer

(W)


Inle

t and

out

let

tem

pera

ture

(∘C)


050100150200250300350

20

25

30

35

40

45

1 12 23 34 45 56 67 78 89 100

111

122

133

144

166

177

188

199

210

221

232

243

254

Time (min)


CPU usage ()

Power (W)



Inle

t and

out

let

tem

pera

ture

(∘C)

CPU

usa

ge (

)po

wer

(W)






Amax inlet 26 to 27




020406080100120

2830323436384042

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)







119894

received le 0


020406080100120

2830323436384042

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)
















1


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)



3 This











(Celsius)





3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)



Out

let t

empe

ratu

re (∘

C)






6 Conclusion


3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


Out

let t

empe

ratu

re (∘






Acknowledgment


References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






2 Δ119879119894gt 0

3 ((119879119894


received) le 0 or (119879119894



5 120583 asymp 0

6 Δ119879119894asymp 0


8 ((119879119894



9 ((119879119894


received) le 0

10 (Δ119879119894asymp 0) and (120583 asymp 0)


12 (Δ119879119894gt 0) and (120583 gt 0)


3


0


1


2


1198783in set 119878

3



(3) 119878119895

TTS rarr 119879119895


TTS rarr 119879119894


0


1198783




(8) 119878119896

TTS rarr 119879119896


TTS rarr 119879119894


119896

TTS rarr 119905119896


1198783rarr 119905119894


1






(14) 119878119898

TTS rarr 119879119898


TTS rarr 119879119894


119898

TTS rarr 119905119898


1198783rarr 119905119894

maxCPU

(16) 119878119898

TTS rarr 119905119898


1198783rarr 119905119894

minCPU



1198783



1198780

(21) End

Algorithm 1


050100150200250300350

202530354045

1 12 23 34 45 56 67 78 89 100

111

122

133

144

155

166

177

188

199

210

221

232

243

254

Time (min)



CPU

usa

ge (

)po

wer

(W)


Inle

t and

out

let

tem

pera

ture

(∘C)


050100150200250300350

20

25

30

35

40

45

1 12 23 34 45 56 67 78 89 100

111

122

133

144

166

177

188

199

210

221

232

243

254

Time (min)


CPU usage ()

Power (W)



Inle

t and

out

let

tem

pera

ture

(∘C)

CPU

usa

ge (

)po

wer

(W)






Amax inlet 26 to 27




020406080100120

2830323436384042

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)







119894

received le 0


020406080100120

2830323436384042

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)
















1


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)



3 This











(Celsius)





3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)



Out

let t

empe

ratu

re (∘

C)






6 Conclusion


3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


Out

let t

empe

ratu

re (∘






Acknowledgment


References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




050100150200250300350

202530354045

1 12 23 34 45 56 67 78 89 100

111

122

133

144

155

166

177

188

199

210

221

232

243

254

Time (min)



CPU

usa

ge (

)po

wer

(W)


Inle

t and

out

let

tem

pera

ture

(∘C)


050100150200250300350

20

25

30

35

40

45

1 12 23 34 45 56 67 78 89 100

111

122

133

144

166

177

188

199

210

221

232

243

254

Time (min)


CPU usage ()

Power (W)



Inle

t and

out

let

tem

pera

ture

(∘C)

CPU

usa

ge (

)po

wer

(W)






Amax inlet 26 to 27




020406080100120

2830323436384042

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)







119894

received le 0


020406080100120

2830323436384042

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)
















1


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)



3 This











(Celsius)





3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)



Out

let t

empe

ratu

re (∘

C)






6 Conclusion


3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


Out

let t

empe

ratu

re (∘






Acknowledgment


References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




020406080100120

2830323436384042

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)
















1


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 13 25 37 49 61 73 85 97 109

121

133

145

157

169

181

193

205

217

229

241

253

265

277

289

301

313

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)


0

20

40

60

80

100

120

32

34

36

38

40

42

44

1 10 19 28 37 46 55 64 73 82 91 100

109

118

127

136

145

154

163

172

181

190

199

208

217

226

235

244

253

262

271

280

289

298

307

316

325

Time (min)


CPU usage ()

Out

let t

empe

ratu

re (∘

C)



CPU

usa

ge (

) in

let

tem

pera

ture

(∘C)



3 This











(Celsius)





3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)



Out

let t

empe

ratu

re (∘

C)






6 Conclusion


3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


Out

let t

empe

ratu

re (∘






Acknowledgment


References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in









(Celsius)





3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)



Out

let t

empe

ratu

re (∘

C)






6 Conclusion


3032343638404244

1 11 21 31 41 51 61 71 81 91 101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

251

261

271

281

291

301

311

321

Time (min)


Out

let t

empe

ratu

re (∘






Acknowledgment


References







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in







































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



Documents

Research Article Minimizing Thermal Stress for Data Center ...downloads.hindawi.com/journals/tswj/2014/684501.pdf · Research Article Minimizing Thermal Stress for Data Center Servers