Integrating Adaptation Mechanisms Using Control Theory Centric Architecture Models: A Case Study

Integrating Adaptation Mechanisms
Using Control Theory Centric
Architecture Models: A Case Study
ICAC’14, Philadelphia, 18.6.2014
Philippe Collet Romain Rouvoy
University Lille 1 / LIFL 
INRIA Lille 
France
Université Nice Sophia
Antipolis I3S - CNRS UMR 7271 
France
University Lille 1 / LIFL 
INRIA Lille 
France
Filip Křikava

ICAC’14, 18.6.2014
Integrating Adaptation Mechanisms Using Control Theory Centric
SELF-ADAPTIVE SOFTWARE SYSTEMS, FEEDBACK CONTROL
• Increasing complexity of software systems, uncertain operating environment
• A shift towards self-adaptive software systems
• systems adjusting themselves in accordance with higher-level goals
• Feedback control loops (FCLs) provide a generic mechanism for engineering
self-adaptive software systems
Monitoring Reconﬁguration
Decision Making
sensors effectors
Target System
events actions
measures decisions
system outputs
(response time,
utilization, throughput)
system control inputs
(queue size, scheduling
policy, concurrency level)
achieve externally specified goals

ICAC’14, 18.6.2014
• Engineering self-adaptive software systems is challenging
• Example: web server self-optimization (MC based on MEM)
CHALLENGES
FCL control challenges
control engineers
FCL integration challenges
software engineers
2500
0
0.5
1
CPU
0
0.5
1
MEM
0
10
20
KA
0 500 1000 1500 2000 2500
0
500
1000
Time (s)
MC
0
0.5
1
CPU
0
0.5
1
MEM
0
10
20
KA
0 500 1000 1500 2000 2500
0
500
1000
Time (s)
MC
(b) SISO data, (c) MIMO data
MC varying
Experimental input-output data used for system ID.
1 = 0.485 · MEMk + 3.63 £ 10°4
· MCk (3)
when performing two separate SISO identifications, it is as-
ng parameters do not interact. However, from Figure 7(b), we
wave has a significant effect on CPU. Hence, our assumption is
he need for MIMO techniques in system identification.
fication
Apache MIMO models shown in Figure 3(b) and Figure 3(c).
del requires simultaneously varying both KA and MC in order to
2500
0
0.5
1
CPU
0
0.5
1
MEM
0
10
20
KA
0 500 1000 1500 2000 2500
0
500
1000
Time (s)
MC
0
0.5
1
CPU
0
0.5
1
MEM
0
10
20
KA
0 500 1000 1500 2000 2500
0
500
1000
Time (s)
MC
(b) SISO data, (c) MIMO data
MC varying
Experimental input-output data used for system ID.
1 = 0.485 · MEMk + 3.63 £ 10°4
· MCk (3)
when performing two separate SISO identifications, it is as-
ng parameters do not interact. However, from Figure 7(b), we
wave has a significant effect on CPU. Hence, our assumption is
he need for MIMO techniques in system identification.
fication
Apache MIMO models shown in Figure 3(b) and Figure 3(c).
del requires simultaneously varying both KA and MC in order to
ween the parameters. As before, discrete sine waves are used.
Controller
Apache
Web
Server
MEMMCMEM⇤
0
0.5
1
CPU
0
0.5
1
MEM
0
10
20
KA
0 500 1000 1500 2000 2500
0
500
1000
Time (s)
MC
0
0.5
1
CPU
0
0.5
1
MEM
0
10
20
KA
0 500 1000 1500 2000 2500
0
500
1000
Time (s)
MC
0
0.5
1CPU
0
0.5
1
MEM
0
10
20
KA
0 500 1000 1500 2000 2500
0
500
1000
Time (s)
MC
(a) SISO data, (b) SISO data, (c) MIMO data
KA varying MC varying
Figure 7: Experimental input-output data used for system ID.
MEMk+1 = 0.485 · MEMk + 3.63 £ 10°4
· MCk (3)
As mentioned earlier, when performing two separate SISO identifications, it is as-
umed that the two tuning parameters do not interact. However, from Figure 7(b), we
an see that the MC sine wave has a significant effect on CPU. Hence, our assumption is
valid. This motivates the need for MIMO techniques in system identification.
.2.2 MIMO Identification
Next, we consider the Apache MIMO models shown in Figure 3(b) and Figure 3(c).
entifying a MIMO model requires simultaneously varying both KA and MC in order to
apture interactions between the parameters. As before, discrete sine waves are used.
owever, their frequencies should be relatively prime so that good coverage of the input
pace is obtained (as in Figure 6).
Figure 7(c) plots the results of the MIMO identification experiments. Using this data,
first-order linear MIMO model is constructed, as shown in Equation 4. Note that A
nd B are matrices instead of scalars. The matrix approach provides a way to model
e combined effect of tuning parameters on utilizations. Specifically, all elements of
e second matrix (B) are non-zero (although the effect of KA on MEM is much less than
s effect on CPU).
CPUk+1
MEMk+1
∏
=
∑
0.537 °0.109
°0.0256 0.630
∏
·
∑
CPUk
MEMk
∏
+
∑
°84.5 4.39
°2.48 2.81
∏
£10°4
·
∑
KAk
MCk
∏
(4)
.3 Model Evaluation
Two model evaluations are done. The first is one-step prediction in which the value
a metric at time k + 1 is predicted based on measured values at time k. The second
multi-step prediction in which the value at k + 1 is predicted based on the prediction
time k. Throughout, we focus on CPU since the low variability of MEM makes it
latively easy to predict.
Figure 8 plots predicted versus measured values for one-step predictions of CPU. A
erfect model would have all observations (the diamonds) on the line of unit slope. Part
) plots the SISO results using the SISO data. The fit is quite good; R2
= 0.934. Part
When |a| < 1, the system is stable and the a terms will remain bounded
s of k. As noted above, the model can predict steady-state values of the output
stant inputs u by finding fixed-points of Equation 5.
er dynamic property that can be determined by the pole is speed of response.
proaches zero, the speed of response to changes in the input becomes faster. If
en yk+1 = b · uk. Hence, it takes one time step for the output, y, to converge
nges in the input. As |a| approaches one, it takes more time steps for the output
ge to a steady-state after changes in the input. This can be seen in Figure 9 in
CP U = 0.595 > 0.485 = aMEM , and MEM settles faster than CPU.
nal property that can be determined by the pole is whether the system will
n an oscillatory manner to changes in the input. If a < 0, then the behavior
stem will be oscillatory due to the aj
terms in Equation 5; when j is even, the
ositive, and when j is odd, the term is negative.
vector case, the scalar parameters a and b in Equation 5 are replaced by matri-
d B. The solution to this vector time-series equation is similar to Equation 5.
se, the response of the system is governed by the eigenvalues of A, which are
in the vector case. Note that in the vector case, the poles can be complex (with
and imaginary parts). The presence of an imaginary part can cause oscillations
he real part is positive. With multiple poles, the “slowest” or dominant poles
se whose magnitude is closest to one) determine settling times. The poles of
O model of Equation 4 are 0.513 and 0.654, indicating that this model predicts
response than that predicted by the SISO model.
trol Design
ection describes our methodology for designing feedback controllers and ap-
o Apache for the architectures shown in Figure 3. Section 4.1 introduces the
logy. Section 4.2 applies this methodology to designing SISO controllers, and
4.3 to MIMO design with LQR. Section 4.4 discusses controller robustness to
n workloads.
esign Methodology
ethodology for controller design uses a proportional integral (PI) controller.
of its robustness, PI control is widely used in mechanical engineering and
ontrol. PI control operates according to the following control law
uk = KP ek + KI
k°1X
j=1
ej, (6)
is a tuning parameter, and ek = rk ° yk is the control error. For the Apache
rk is the desired policy (CPU§
and/or MEM§
), and yk is the measured metric
d/or MEM). For SISO control, the policies (CPU§
, MEM§
) are considered indi-
for MIMO control they are considered together.
trol has two parameters: KP , the proportional gain, and KI, the integral gain.
of thumb, the proportional term is used to increase the speed of response and
[Hellerstein et al., 2004]
• Design decision mechanism
• data collection
• model construction and evaluation
• controller design
API
LOGS
CONFIG FILES
COMMANDS
AOP
PROFILER
• Prepare experimental environment
• identify system outputs (sensors)
• identify system control inputs (effectors)
// Type definitions ---------------------------------------
typedef struct { // Positional ISA PID
float K,Ti,Td,N,b,c,CSmin,CSmax,Ts; // parameters
float Io,Do; // state vars
float CSo; // for locks
float SPo,PVo; // input mem
float SP,PV,TR,CS; // I/O vars
char TS,Fp,Fm,HI,LO;
} ISAPIDpos_DATA;
typedef struct { // Incremental ISA PID
float CSo,Do; // state vars
} ISAPIDinc_DATA;
typedef struct { // 1st order t.f. w/o zeroes
float mu,T,Ts; // parameters
float xo; // state vars
float u,y; // I/O vars
} TF1P0Z_DATA;
// Dynamic systems (init and run) -------------------------
void isapidpos_coldinit(ISAPIDpos_DATA *pd,
float K, float Ti, float Td,
float N, float b, float c,
float CSmin, float CSmax, float Ts,
char TS, char Fp, char Fm)
{
pd->K = K; pd->Ti = Ti; pd->Td = Td;
pd->N = N; pd->b = b; pd->c = c;
pd->CSmin = CSmin; pd->CSmax = CSmax; pd->Ts = Ts;
pd->TS = TS; pd->Fp = Fp; pd->Fm = Fm;
pd->Io = 0.0; pd->Do = 0.0; pd->SPo = 0.0;
pd->PVo = 0.0; pd->CS = 0.0;
pd->HI = 0; pd->LO = 0;
}
void isapidpos(ISAPIDpos_DATA *pd)
{
float DSP,DPV,P,I,D;
if (pd->TS==0) { // Automatic mode
DSP = pd->SP-pd->SPo;
DPV = pd->PV-pd->PVo;
P = pd->K*(pd->b*pd->SP-pd->PV);
I = pd->Io+pd->K*pd->Ts/pd->Ti*(pd->SP-pd->PV);
D = (pd->Td*pd->Do+pd->K*pd->N*pd->Td
*(pd->c*DSP-DPV))/(pd->Td+pd->N*pd->Ts);
pd->CS = P+I+D;
if ((pd->CS>pd->CSo & pd->Fp)
|(pd->CS<pd->CSo & pd->Fm))
pd->CS = pd->CSo;
}
else { // Tracking mode (can override locks)
pd->CS = pd->TR;
D = 0; }
// Antiwindup
if (pd->CS>pd->CSmax) pd->CS = pd->CSmax;
if (pd->CS<pd->CSmin) pd->CS = pd->CSmin;
pd->Io = pd->CS-P-D;
// Saturation signalling
} ISAPIDpos_DATA;
} ISAPIDinc_DATA;
} TF1P0Z_DATA;
{
pd->N = N; pd->b = b; pd->c = c;
pd->Io = 0.0; pd->Do = 0.0; pd->SPo = 0.0;
pd->PVo = 0.0; pd->CS = 0.0;
pd->HI = 0; pd->LO = 0;
}
{
pd->CS = P+I+D;
pd->CS = pd->CSo;
}
pd->CS = pd->TR;
D = 0; }
// Antiwindup
} ISAPIDpos_DATA;
} ISAPIDinc_DATA;
} TF1P0Z_DATA;
{
pd->N = N; pd->b = b; pd->c = c;
pd->Io = 0.0; pd->Do = 0.0; pd->SPo = 0.0;
pd->PVo = 0.0; pd->CS = 0.0;
pd->HI = 0; pd->LO = 0;
}
{
pd->CS = P+I+D;
pd->CS = pd->CSo;
}
pd->CS = pd->TR;
D = 0; }
// Antiwindup
• Implementation
• integration into target system
• consisten monitoring
• coordinated reconfiguration

ICAC’14, 18.6.2014
INTEGRATING ADAPTATION MECHANISMS
• Adhoc Implementations
• Extensive handcrafting of non-trivial code
• Low level of FCL abstraction
• Low FCL visibility
• Limited verification and reasoning
• Giving raise to accidental complexities
• Reusing and adapting existing work
• Often target specific types of adaptation problems [Bertran’12]
• Require the use of certain adaptation mechanism [Garlan’04]
• Applicable to single domain [Rouvoy’08] or technology [Asadollahi’09]
• Do not support remoting or complex control schemes [Adamczyk’08,
Gorton’06]
• Overall lack of supporting control theory based controller
• Limiting their applicability
API
LOGS
CONFIGURATIONS
COMMANDS
AOP
PROFILER

ICAC’14, 18.6.2014
OVERVIEW
Monitoring Reconﬁguration
Decision Making
sensors effectors
Target System
events actions
measures decisions
Controller Target System
Transduc
er
+
-
=
mX
i=1
1
d
=
m
d
⇥(N) =
µ =
m
d
µ
d =
m
(N)µ
API
LOGS
CONFIGURATIONS
COMMANDS
AOP
PROFILER
Feedback Control Definition Language1
The ZNN.COM case study2
Systematic integration of self-adaptive mechanisms into software
systems through control theory centric architecture models
Generality Visibility Composition

ICAC’14, 18.6.2014
1Feedback Control Definition Language

ICAC’14, 18.6.2014
FEEDBACK CONTROL DEFINITION LANGUAGE - IN A NUTSHELL
Controller
Apache
Web
Server
MEMMCMEM⇤
Apache adaptation example - adjusts the maximum number of connections
to be processed simultaneously (MC) based on the difference between
reference (MEM*) and measured memory usage (MEM) [Hellerstain’04].

ICAC’14, 18.6.2014
FEEDBACK CONTROL DEFINITION LANGUAGE (FCDL) - IN A NUTSHELL
Controller
Apache
Web
Server
MEMMCMEM⇤
Apache adaptation example - adjusts the maximum number of connections
to be processed simultaneously (MC) based on the difference between
reference (MEM*) and measured memory usage (MEM) [Hellerstain’04].
(effector)
(controller)
(processor)
(sensor)
memUsage
: ProcessMem
controller
: IController
mcConf
: SetApacheConf
out outputin input
out output
scheduler
: PeriodTrigger
out output
in input
in input
MEM
MEM
MC
-refMem
-kI
ps -o pmem= $(cat /var/run/httpd.pid)
/etc/apache2/httpd.conf
/etc/init.d/apachectl
u(t) = u(k 1) + Kie(k)
- initialPeriod
- httpdConfPath- pid
(target system)
↵ = h self; + (input); * (output?) i
(provided sensor) (provided effector)
in input
setRefMemrefMem
out output
MEM⇤ MEM⇤
• Domain-specific modeling language
• Network of Adaptive Elements (AEs)
• actor-like element
• push/pull communication
• properties
• specifies implementations
• supports reflection
• defines interaction contracts

ICAC’14, 18.6.2014
THE ZNN.COM CASE STUDY
2

ICAC’14, 18.6.2014
THE ZNN.COM CASE STUDY
http://www.hpi.uni-potsdam.de/giese/public/selfadapt/exemplars/model-problem-znn-com/
• A news service model
• Web-based N-tier client-server
• Originally used for Rainbow evaluation [Cheng’09]
• Acknowledged SEAMS case-study
• Partially used by others (e.g., DYNAMICO, Zenshin, AdaptCases)
!
• Objective
• “To serve the content within reasonable response time and quality in the
event of traffic spikes caused by highly popular news.“
!
• Challenges
• content adaptation (e.g. serving reduced content quality),
• service differentiation (e.g. prioritizing premium customers),
• infrastructure adaptation (e.g. adapting the size of the server pool)

ICAC’14, 18.6.2014
Idea: service time = fixed overhead + data-size dependent overhead
QoS management control of web servers by content delivery adaptation
[Abdelzaher et al., 1999, 2002]
...
serve from
tree #2
serve from
tree #1
Rejection
Level
Minimum Content Full Content
Goal: maintain server load around some pre-set value
/1/photo.jpg
/2/photo.jpg
/photo.jpg
full quality
degraded quality
normal load
overload
Prerequisite: preprocessed content (different quality and size)
ZNN.COM - LOCAL CONTENT DELIVERY ADAPTATION
Generality Visibility Composition
Using FCDL

ICAC’14, 18.6.2014
Compute severity of adaptation (G)
Compute the requests rate (R), bandwidth (W) and utilization (U)
Compute the number of requests (r) and size of responses (w)1
2
3

ICAC’14, 18.6.2014
Compute the number of requests (r) and size of responses (w)1
accessLog
: FileTailer
out lines
ﬁle=/var/log/apache2/access.log
access_log
active sensor FileTailer {
push out port lines: String
!
property file: String
}
processor Accumulator {
push in port input: long
pull out port sum: long
}
processor AccessLogParser {
push in port lines: String
push out port size: int
push out port requests: int
}
Generality
Visibility
requestCounter
: Accumulator
responseSizeCounter
: Accumulator
in input
out sum out sum
in lines
out sizeout requests
accessLogParser
: AccessLogParser

ICAC’14, 18.6.2014
access_log
active processor PeriodTrigger<T> {
pull in port input: T
push out port output: T
!
property initialPeriod = 10.seconds
}
requestCounter
: Accumulator
responseSizeCounter
: Accumulator
scheduler
: PeriodTrigger
loadMonitor
: LoadMonitor
in input
out sum out sum
in requests in size
out utilization
in input
out output
initialPeriod=10s
accessLog
: FileTailer
in lines
out lines
ﬁle=/var/log/apache2/access_log
accessLogParser
: AccessLogParser
Compute the requests rate (R), bandwidth (W) and utilization (U)2
Generality
Visibility

ICAC’14, 18.6.2014
content_treeaccess_log
Compute severity of adaptation (G)3
controller IController { 
in push port input: double 
out push port output: double 
!
property KI: double
property reference: double
property loBnd: double
property upBnd: double
}
utilController
: IController
in input
requestCounter
: Accumulator
responseSizeCounter
: Accumulator
scheduler
: PeriodTrigger
loadMonitor
: LoadMonitor
in input
out sum out sum
in requests in size
out utilization
in input
out output
in input
out output
initialPeriod=10s
adaptor
: ContentAdaptor
in contentTree
accessLog
: FileTailer
in lines
out lines
accessLogParser
: AccessLogParser
KI=?
reference=?
Generality
Visibility

ICAC’14, 18.6.2014
Composition
composite ApacheQOS {
!
feature accessLog = new FileTailer {
file = “/var/log/apache2/access_log”
}
!
feature accessLogParser = new AccessLogParser
feature requestCounter = new Accumulator
feature responseSizeCounter = new Accumulator
feature loadMonitor = new LoadMonitor
feature scheduler = new PeriodTrigger<Double>
feature utilController = new IController {
reference = 0.8
}
feature adaptor = new ContentAdaptor
!
connect accessLog.lines to
accessLogParser.lines
connect accessLogParser.size to
responseSizeCounter.input
connect accessLogParser.requests to
requestsCounter.input
connect requestCounter.output to
loadMonitor.requests
connect responseSizeCounter.output to
loadMonitor.size
connect loadMonitor.utilization to
scheduler.input
connect scheduler.output to
utilController.utilization
connect utilController.contentTree to
adaptor.contentTree
}
Complete model
ApacheQOS
utilController
: IController
in input
requestCounter
: Accumulator
responseSizeCounter
: Accumulator
scheduler
: PeriodTrigger
loadMonitor
: LoadMonitor
in input
out sum out sum
in requests in size
out utilization
in input
out output
in input
out output
system
layer
control
layer
initialPeriod=10s
adaptor
: ContentAdaptor
in contentTree
accessLog
: FileTailer
in lines
out lines
accessLogParser
: AccessLogParser
KI=?
reference=?
Generality Visibility

ICAC’14, 18.6.2014
Generality
Visibility
ApacheQOS
control
: QOSControl
in contentTree
apache
: ApacheWebServer
in requests
in size
out requests
out size
out contentTree
control
layer
system
layer
QOSControl utilController
: IController
utilization
: UtilizationMonitor
out contentTree
scheduler
: PeriodTrigger
out utilization
in input out output
in input
out outputin requests in size
in requests in size
adaptor
: ContentAdaptor
outsize
in contentTree
out size
outrequests
out requests
incontentTree
ApacheWebServer
accessLog
: FileTailer
accessLogParser
: AccessLogParser
in lines
out lines
Composition

ICAC’14, 18.6.2014
ZNN.COM - ADAPTIVE CONTROL
• Using the reflection support for adaptive control
AdaptiveApacheQOS
out utilization
controllayer
meta-control
layer
systemlayer
in contentTree
in requests
in size
out requests
out size
out contentTree
provided in KI
control
: QOSControl
out KI
in utilization
in contentTree
KI
server
: ApacheWebServer
adaptiveControl: AdaptiveControl

ICAC’14, 18.6.2014
ILLUSTRATION - DISTRIBUTED CONTENT DELIVERY ADAPTATION
• QoS management control of a pool of web servers using content
delivery adaptation.
• Load balancer schedules requests to a server with highest QoS.
Znn.com
Client 1
Client 2
Client N
Forwards
requests to s
Server cluster
...
Load Balancer
...
S
G1
G2
Gn
s 2 S such that Gs = max (G1, . . . , Gn)

ICAC’14, 18.6.2014
LocalApacheQOS
out contentTree
contentTreeWithHost
: Tuple2
hostname
: Hostname
Si
out output
out hostname
in input2
in input1
in input
Gi
contentTreePub
: EventBusPublisher
server
: ApacheQOS
(Si, Gi)
scheduler
: PeriodTrigger
in input
out output
(Si, Gi)
contentTree
: Queue
in input
out output
Gi
outcontentTree
ApacheQOS
in contentTree
in requests
in size
out requests
out size
out contentTree
server
: ApacheWebServer
control
: QOSControl
LoadBalancerControl
contentTree
: MapStore
ctrl
: LoadBalancerController
contentTreeMin
: MapMaxKey
lbForwardHost
: FileWriter
out output out outputin input
in input
Siout output in value
in hostname
out hostname
8
><
>:
S1 : G1
...
Sn : Gn
9
>=
>;
contentTreeSub
: EventBusSubscriber
scheduler
: PeriodTrigger
in input
out output
Si Si
(Si, Gi)
Load Balancer
si
ILLUSTRATION - DISTRIBUTED CONTENT DELIVERY ADAPTATION

ICAC’14, 18.6.2014
ILLUSTRATION - RESOURCE MANAGEMENT
ServerPoolControl
contentTree
: MapStore
serverPoolController
: IController
contentTreeAvg
: MapAvgValue
clusterManager
: ClusterManager
out output out output
n
in input
in input
out output in numOfServers
in input
out output
8
><
>:
S1 : G1
...
Sn : Gn
9
>=
>;
scheduler
: PeriodTrigger
in input
out output
(Si, Gi)
G G
contentTreeSub
: EventBusSubscriber
Load Balancer
…
s1
sn
G1
Gn
n = n + KIE = n + KI(G⇤
G)
n

ICAC’14, 18.6.2014
ZNN.COM - SYSTEM IDENTIFICATION
• Support for FCL design - black-box modeling
• Open control loops for data collection
SystemIdentiﬁcation
out output
log
: FileWriter
in value
out utilization
in requests
in size scheduler
: PeriodTrigger
in input out output
sample: Tuple2
in input2
in input1
gQueue: Queue
out output
out requests
out size
in contentTree
contentTree: SineWaveserver: ApacheWebServer
(U, G) (U, G)
utilization
: UtilizationMonitor
in input

ICAC’14, 18.6.2014
Modeling
VeriﬁcationCode Generation
• Reference implementation of FCDL based on Eclipse Modeling Framework
• Eclipse IDE-based prototype to facilitate the use of FCDL - ACTRESS
IMPLEMENTATION
accessLogParser accessLog adaptor
apache control
requestCounter
responseSizeCounter
loadMonitor
utilization
scheduler utilController
ApacheQOS
ACTRESS Runtime
actor
actor with
event listener
composite
actor
containment
message
passing
ApacheQOS utilController
: IController
in input
requestCounter
: Accumulator
responseSizeCounter
: Accumulator
scheduler
: PeriodTrigger
loadMonitor
: LoadMonitor
in input
out sum out sum
in requests in size
out utilization
in input
out output
in input
out output
out requests
out size
initialPeriod=10s
in contentTree
server
: ApacheWebServer
KI=?
reference=?
• Model well-formedness
• User-defined structural constraints
(OCL)
• Architecture verification
(completeness, consistency,
determinacy)
• Temporal properties (PROMELA/SPIN)
• xFCDL (Extended FCDL)
• Textual DSL for authoring
FCDL models
• Modularity, Java
interoperability, Xbase
• Eclipse IDE support
• ACTRESS runtime
• based on Java/AKKA

ICAC’14, 18.6.2014
Conclusions
3

ICAC’14, 18.6.2014
SUMMARY
• Combining self-adaptive software systems with principles of MDE to provide
systematic and tooled approach for integrating control mechanisms into
software systems.
!
• A proof of concept implementation and tools facilitating the language use
including modeling, code synthesis and verification support.

ICAC’14, 18.6.2014
FUTURE WORK
• Address ACTRESS limitations - a new MPS-based implementation
• More robust FCDL - support data units, input output assertions
• ZNN.COM challenge
• From a case study to a benchmark
• Easily reusable (docker packaging)
• INRIA Lille Non-A team in charge of a new controller
• Evaluation agains two base lines
• Apache mod_proxy
• Amazon elastic load balancer

Thank you
Filip Křikava
filip.krikava@inria.fr

Integrating Adaptation Mechanisms Using Control Theory Centric Architecture Models: A Case Study

More Related Content

What's hot (20)

Viewers also liked (11)

Similar to Integrating Adaptation Mechanisms Using Control Theory Centric Architecture Models: A Case Study (20)

Recently uploaded (20)

Integrating Adaptation Mechanisms Using Control Theory Centric Architecture Models: A Case Study