Wireless example Guidelines References
Design of ExperimentsOption PDES
Jean-Marc Vincent1
Laboratoire LIG{Jean-Marc.Vincent}@imag.fr
1Université Joseph FourierÉquipe-Projet MESCAL
Some elements for the design of experiments : an example
1 / 21Design of Experiments
Wireless example Guidelines References
Wireless environment
Environment
Wireless DevicesLaptop computers,
Personal digital assistants (PDAs),
Mobile phones, ...
Wireless Ad-Hoc NetworkWIFI,
Bluetooth, ...
PrincipleDevices share services and collaborate to maintain the community.
General distributed system problem→ Dynamical control of the architecture.→ State of the system observation and distributed decision process.
2 / 21Design of Experiments
Wireless example Guidelines References
Wireless environment
Environment
Wireless DevicesLaptop computers,
Personal digital assistants (PDAs),
Mobile phones, ...
Wireless Ad-Hoc NetworkWIFI,
Bluetooth, ...
PrincipleDevices share services and collaborate to maintain the community.
General distributed system problem→ Dynamical control of the architecture.→ State of the system observation and distributed decision process.
2 / 21Design of Experiments
Wireless example Guidelines References
Operating systems on a wireless network
Dynamic Architecture
Heterogeneity of devices
Behavior of wireless devices→ connections / disconnectionsBehavior of wireless network
Unreliability of communicationsVariability of latencies
Solution to maintain the consistency of the community
Design and adapt distributed algorithms ;
to make some distributed decisions.
(consensus, election, atomic broadcast, group membership,...)
3 / 21Design of Experiments
Wireless example Guidelines References
The Consensus Problem
The impossibility of Fisher, Lynch & Paterson [Fischer-Lynch-Paterson 85]
Some approaches to circumvent this impossibility result :
Probabilistic algorithms [Canetti-Rabin 93]
Self-Stabilizing algorithms [Tixeuil 00]
An approach with partial synchrony [Dwork-Lynch-Stockmeyer 88]
....
⇒ The selected approach :
Consensus + unreliable failure detectors [Chandra-Toueg 96]
Interest :Dynamicity analysis of the environment is concentrated only inside failure detectors.
Objective :Guarantee a quality of service for failure detectors.
4 / 21Design of Experiments
Wireless example Guidelines References
The Consensus Problem
The impossibility of Fisher, Lynch & Paterson [Fischer-Lynch-Paterson 85]
Some approaches to circumvent this impossibility result :
Probabilistic algorithms [Canetti-Rabin 93]
Self-Stabilizing algorithms [Tixeuil 00]
An approach with partial synchrony [Dwork-Lynch-Stockmeyer 88]
....
⇒ The selected approach :
Consensus + unreliable failure detectors [Chandra-Toueg 96]
Interest :Dynamicity analysis of the environment is concentrated only inside failure detectors.
Objective :Guarantee a quality of service for failure detectors.
4 / 21Design of Experiments
Wireless example Guidelines References
Unreliable Failure Detectors
Principle :For each remote device, build an estimation of the global state.
Local view of the global system.→ List of suspected devices
Properties
Accuracy : a correct process should not be suspected
Completeness : an incorrect process should be suspected
Quality of serviceQuality of information and reactivity
false suspicion rate = function(reactivity)
- false suspicion type 1 : correct process suspected- false suspicion type 2 : crashed process not suspectedRisk analysis
Implementation
5 / 21Design of Experiments
Wireless example Guidelines References
Unreliable Failure Detectors
Principle :For each remote device, build an estimation of the global state.
Local view of the global system.→ List of suspected devices
Properties
Accuracy : a correct process should not be suspected
Completeness : an incorrect process should be suspected
Quality of serviceQuality of information and reactivity
false suspicion rate = function(reactivity)
- false suspicion type 1 : correct process suspected- false suspicion type 2 : crashed process not suspectedRisk analysis
Implementation
SuspicionSuspicion
No
ReceiptHeartbeat
Timeout
5 / 21Design of Experiments
Wireless example Guidelines References
Failure Detectors Implementation
Need of information on remote devices
Export local data
Collect and analyse data coming from remote devices
Informations Export
→ Broadcast of information by anticipation (Heartbeatmecanism)
ConsultationInfo publication
Information
parameters )&
Middleware Interface
moduleExport
( policyMW control
Suspicion requestNetwork Interface
Information
List of suspects
Network interface Suspicion request
Importmodule
Consensus interface
parameters )
( policy&
MW control
MiddlewareInterface
List to query
Informations Import
→ Collect information
→ Estimate the state of remote devices
6 / 21Design of Experiments
Wireless example Guidelines References
Failure Detectors Parameters
The running principle :Implemented mecanism : "heartbeat".
� � �� � �� � �� � � � �
� ���
FD ExportProcess
SenderDevice
DeviceReceiver
ProcessFD Import
FD sending delay
FD receipt delayθ
θ θθ
θ
Parameters
Heartbeat sending period.
Estimate function of suspicions (timeout). [Bertier-Marin-Sens 03]
GoalEstimate the timeout values according to the expected quality of service.
7 / 21Design of Experiments
Wireless example Guidelines References
Failure Detectors Parameters
The running principle :Implemented mecanism : "heartbeat".
� � �� � �� � �� � � � �
� ���
FD ExportProcess
SenderDevice
DeviceReceiver
ProcessFD Import
FD sending delay
FD receipt delayθ
θ θθ
θ
Parameters
Heartbeat sending period.
Estimate function of suspicions (timeout). [Bertier-Marin-Sens 03]
GoalEstimate the timeout values according to the expected quality of service.
7 / 21Design of Experiments
Wireless example Guidelines References
Statistical Description
Network
Network
FD ExportProcess
FD ImportProcess
FD sending delay
Network sending delay
Network receipt delay
FD receipt delay
SenderDevice
ReceiverDeviceHB interval (measured)
HB Interval (given)
Variability of HB arrivals
λ0 = emission beat rate
Xi = Heartbeat inter-arrivals.
λ = 1HB period (assumption : few losses λ = λ0.(1− loss probability)
8 / 21Design of Experiments
Wireless example Guidelines References
Statistical Description
False Detection Probabilityθ = suspicion threshold (timeout)φI (θ) = asymptotic false suspicion rate
φI (θ) = λ limn→∞
1
n
n∑i=1
(Xi − θ)+
If the inter-arrivals {Xi} of beats are independent and identically distributed, then :
φI (θ) = λEπ [X − θ]+
where π is the distribution of Xi . (renewal process)
9 / 21Design of Experiments
Wireless example Guidelines References
Independent assumption Model
Variable Sending Delay
Hypothesis : {Xi} : renewal process (iid)
Model suspicion rate
no information on variance Exponential φI (θ) = e−λθ
low variation coefficient Erlang(k,kλ) φI (θ) = e−kλθPk (λθ)
high variation coefficient Pareto(α) φI (θ) = 1(1+ θ
α−2 )α−2
Suspicion probability related to reactivity
Erlang model : Pareto model :
10 / 21Design of Experiments
Wireless example Guidelines References
Independent assumption Model
Variable Sending Delay
Hypothesis : {Xi} : renewal process (iid)
Model suspicion rate
no information on variance Exponential φI (θ) = e−λθ
low variation coefficient Erlang(k,kλ) φI (θ) = e−kλθPk (λθ)
high variation coefficient Pareto(α) φI (θ) = 1(1+ θ
α−2 )α−2
Suspicion probability related to reactivity
Erlang model :
k=1 (exponential model)
k=2
k=3
k=4k=5
False suspicion rate
1e−05
1e−04
0.001
0.01
0.1
0 1 2 3 4 5 6 θ
1
Pareto model :
σ=
σ=2
3
σ=1.1 0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
2 4 6 8 10 12 14 16 18 θ
10 / 21Design of Experiments
Wireless example Guidelines References
Contention on receiver (1)
Variability of Heartbeat Arrivals
Depends on the type of receiver (Laptop or PDA)
+ Correlation between inter-beats arrival periods
Network
FD ImportProcess
ReceiverDevice
variability of delivery
⇒ HB contention on the receiver
Buffer
Network
Service
Delivery to
upper layer
Heart−beats Failure detector
input process : {An}n∈N
service model : {Sn}n∈N
hypothesis : deterministic arrivals
→ GI/M/1 queue
11 / 21Design of Experiments
Wireless example Guidelines References
Contention on receiver (2)
Output process of a D/M/1 queue → Suspicion rate computation
Inter-output period : A = 1λ
; β unique solution of β = LA(µ(1− β)) = e−Aµ(1−β)
fZ (x) =
{µ
2−β e−µ(1−β)A((1− β)eµ(1−β)x + e−µx ) if x < A;µ
2−β e−µx (e−µ(1−β)A + (1− β)eµA) if x ≥ A.
Rate of false suspicion : θ > A
φI(θ) =1
A(2− β)µe−µθ(e−µ(1−β)A + (1− β)eµA) θ ≥ A
False suspicion probability, D/M/1 model
12 / 21Design of Experiments
Wireless example Guidelines References
Contention on receiver (2)
Output process of a D/M/1 queue → Suspicion rate computation
Inter-output period : A = 1λ
; β unique solution of β = LA(µ(1− β)) = e−Aµ(1−β)
fZ (x) =
{µ
2−β e−µ(1−β)A((1− β)eµ(1−β)x + e−µx ) if x < A;µ
2−β e−µx (e−µ(1−β)A + (1− β)eµA) if x ≥ A.
Rate of false suspicion : θ > A
φI(θ) =1
A(2− β)µe−µθ(e−µ(1−β)A + (1− β)eµA) θ ≥ A
False suspicion probability, D/M/1 model
µ=2 µ=10
A=1 A=1 1e−07
1e−06
1e−05
1e−04
0.001
0.01
0.1
1
10
0 1 2 3 4 5 6 7
False probability suspicion
θ 1e−35
1e−30
1e−25
1e−20
1e−15
1e−10
1e−05
1
100000
0 1 2 3 4 5 6 7 θ
False probability suspicion
12 / 21Design of Experiments
Wireless example Guidelines References
Experimental environment
Contexte Expérimental - Latences (2)
Pocket PCs en mode ad-hoc
0
200
400
600
800
1000
1200
1400
0 50 100 150 200 250 300 350 400 450 500latence en ms
sans economie d’energieeconomie d’energie
Taille échantillon :5000 mesures! ping (délai : 1 s)
{Corine.Marchand, Jean-Marc.Vincent}@imag.fr – CFSE’2003 – Detecteurs de Defaillances et Qualite de Service dans un Reseau Ad-Hoc Heterogene – 10
13 / 21Design of Experiments
Wireless example Guidelines References
Experimental environment
14 / 21Design of Experiments
Wireless example Guidelines References
Contexte Expérimental - Latences (3)
! Grand nombre de paramètres
Facteurs influentsretenus :
Distance
Nb obstacles
Nb entités
Charge réseau
Type émetteur
Type récepteur
Economie NRJ
{Corine.Marchand, Jean-Marc.Vincent}@imag.fr – CFSE’2003 – Detecteurs de Defaillances et Qualite de Service dans un Reseau Ad-Hoc Heterogene – 11
15 / 21Design of Experiments
Wireless example Guidelines References
Contexte Expérimental - Latences (3)
! Grand nombre de paramètres
" Plan d’expérience :! Méthode de Taguchi(à 2 niveaux)
Facteurs influentsretenus :
Distance
Nb obstacles
Nb entités
Charge réseau
Type émetteur
Type récepteur
Economie NRJ
{Corine.Marchand, Jean-Marc.Vincent}@imag.fr – CFSE’2003 – Detecteurs de Defaillances et Qualite de Service dans un Reseau Ad-Hoc Heterogene – 12
15 / 21Design of Experiments
Wireless example Guidelines References
Contexte Expérimental - Latences (3)
Facteurs influentsretenus :
Distance
Nb obstacles
Nb entités
Charge réseau
Type émetteur
Type récepteur
Economie NRJ
{Corine.Marchand, Jean-Marc.Vincent}@imag.fr – CFSE’2003 – Detecteurs de Defaillances et Qualite de Service dans un Reseau Ad-Hoc Heterogene – 13
15 / 21Design of Experiments
Wireless example Guidelines References
Contexte Expérimental - Latences (3)
Facteursprépondérants:
Type récepteur
Charge réseau
Economie NRJ
Type émetteur
Interactions
{Corine.Marchand, Jean-Marc.Vincent}@imag.fr – CFSE’2003 – Detecteurs de Defaillances et Qualite de Service dans un Reseau Ad-Hoc Heterogene – 14
15 / 21Design of Experiments
Wireless example Guidelines References
Experimental environment
Experimental Design :
Devices (same OS, Java) :Architecture 1 : 4 devices (2 Laptops + 2 PDAs)Architecture 2 : 6 devices (2 Laptops + 3 PDAs + 1 sensor
Interconnection : 802.11b ad-hoc network
Experimental duration : 15 min (→ about 10,000 measurements)
HB parameter Settings :
Architecture 1 Architecture 2 Architecture 2
Highly loaded Ideal Setting Perturbed Environment
HB emission period 100 ms 500 ms 500 ms
Timeout none none none
Reception process analysisDensity of the delivery process⇒ timeout tuning
16 / 21Design of Experiments
Wireless example Guidelines References
Experimental environment
Experimental Design :
Devices (same OS, Java) :Architecture 1 : 4 devices (2 Laptops + 2 PDAs)Architecture 2 : 6 devices (2 Laptops + 3 PDAs + 1 sensor
Interconnection : 802.11b ad-hoc network
Experimental duration : 15 min (→ about 10,000 measurements)
HB parameter Settings :
Architecture 1 Architecture 2 Architecture 2
Highly loaded Ideal Setting Perturbed Environment
HB emission period 100 ms 500 ms 500 ms
Timeout none none none
Reception process analysisDensity of the delivery process⇒ timeout tuning
16 / 21Design of Experiments
Wireless example Guidelines References
Highly loaded system
Distribution of the update times :
0
0.002
0.004
0.006
0.008
0.01
0 200 400 600 800 1000
sender: PDA1 - receiver: PDA2
Times between 2 heartbeat receipts
Prob
abili
ty
250 300 350 400 450 500
sender: PDA1 - receiver: laptop2
Prob
abili
ty
0.02
0.015
0.01
0.005
00 50 100 150 200Times between 2 heartbeat receipts
00.0010.0020.0030.0040.0050.0060.0070.0080.009
0 200 400 600 800 1000Times between 2 heartbeat receipts
sender: laptop1 - receiver: PDA2
Prob
abili
ty
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0 50 100 150 200Pr
obab
ility
Times between 2 heartbeat receipts
sender: laptop1 - receiver: laptop2
If timeout value = 200 ms→ Quality of service highly depends on the type of the receiver
17 / 21Design of Experiments
Wireless example Guidelines References
Highly loaded system
Distribution of the update times :
0
0.002
0.004
0.006
0.008
0.01
0 200 400 600 800 1000
sender: PDA1 - receiver: PDA2
Times between 2 heartbeat receipts
Prob
abili
ty
250 300 350 400 450 500
sender: PDA1 - receiver: laptop2
Prob
abili
ty
0.02
0.015
0.01
0.005
00 50 100 150 200Times between 2 heartbeat receipts
00.0010.0020.0030.0040.0050.0060.0070.0080.009
0 200 400 600 800 1000Times between 2 heartbeat receipts
sender: laptop1 - receiver: PDA2
Prob
abili
ty
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0 50 100 150 200Pr
obab
ility
Times between 2 heartbeat receipts
sender: laptop1 - receiver: laptop2
If timeout value = 200 ms→ Quality of service highly depends on the type of the receiver
17 / 21Design of Experiments
Wireless example Guidelines References
"Ideal Setting" Experimentation
Heartbeat Reception Analysis :
0
0.02
0.04
0.06
0.08
0.1
0.12
400 450 500 550 600
Prob
abili
ty
Elapsed times between 2 receipts
sender: laptop1 − receiver: laptop2sender: laptop1 − receiver: pda2sender: pda1 − receiver: pda2
0.14sender: pda1 − receiver: laptop2
Timeout value :
If the timeout value = 2∗ (HB period time)Then, the suspicion rate is around
10−3 if the receiver is a laptop10−2 if the receiver is a PDA
18 / 21Design of Experiments
Wireless example Guidelines References
"Ideal Setting" Experimentation
Heartbeat Reception Analysis :
0
0.02
0.04
0.06
0.08
0.1
0.12
400 450 500 550 600
Prob
abili
ty
Elapsed times between 2 receipts
sender: laptop1 − receiver: laptop2sender: laptop1 − receiver: pda2sender: pda1 − receiver: pda2
0.14sender: pda1 − receiver: laptop2
Timeout value :
If the timeout value = 2∗ (HB period time)Then, the suspicion rate is around
10−3 if the receiver is a laptop10−2 if the receiver is a PDA
18 / 21Design of Experiments
Wireless example Guidelines References
Experimentation in perturbed environment
Perturbation control :An external device is used to generate an external load (ping with 200kbytes/s)
Heartbeat Reception Analysis :
0
0.02
0.04
0.06
0.08
0.1
450 500 550 600 650 700Elapsed Times between 2 heartbeat receipts
sender: Laptop1 − receiver: Laptop2sender: Laptop1 − receiver: PDA2sender: PDA1 − receiver: Laptop2
Prob
abili
tysender: PDA1 − receiver: PDA2
Results :
Long non receiving period for PDA
Some very small delays between HB receipts (after a long waiting time)
⇒ Correlation between succesive waiting times of two HB (bursty receptions)
19 / 21Design of Experiments
Wireless example Guidelines References
Experimentation in perturbed environment
Perturbation control :An external device is used to generate an external load (ping with 200kbytes/s)
Heartbeat Reception Analysis :
0
0.02
0.04
0.06
0.08
0.1
450 500 550 600 650 700Elapsed Times between 2 heartbeat receipts
sender: Laptop1 − receiver: Laptop2sender: Laptop1 − receiver: PDA2sender: PDA1 − receiver: Laptop2
Prob
abili
tysender: PDA1 − receiver: PDA2
Results :
Long non receiving period for PDA
Some very small delays between HB receipts (after a long waiting time)
⇒ Correlation between succesive waiting times of two HB (bursty receptions)
19 / 21Design of Experiments
Wireless example Guidelines References
Guideline (Montgomery)
1 Recognition of and statement of the problemseries of small experimentsfactors screening
2 Selection of the response variable3 Choice of factors, levels, and ranges potential design factors (nuisance factors)
factors controllabilitycause and effects diagram (fishbone diagram)
4 Choice of experimental designrandomization, replication, blockingfactorial planning
5 Performing the experiment6 Statistical analysis of the data7 Conclusions and recommendations
20 / 21Design of Experiments
Wireless example Guidelines References
Références I
Jain, R. (1991), Art of Computer Systems Performance Analysis, JOHNWILEY and SONS.
Lawson, J., ed. (2014), Design and Analysis of Experiments with R,Chapman and Hall/CRC.
Lilja, D. J. (2009), Measuring Computer Performance : A Practitioner’s Guide,Cambridge University Press.
Linder, R. (n.d.), Les plans d’expériences (French Edition), PRESSESECOLE NATIONALE PONTS CHAUSSEES.
Montgomery, D. C. (2009), Design and Analysis of Experiments, 7EInternational Student Version, 7th international student edition edn,John Wiley and Sons Ltd.
21 / 21Design of Experiments