Zahra Abbasi Adel Dokhanchi

chirpskulkInternet και Εφαρμογές Web

3 Νοε 2013 (πριν από 3 χρόνια και 9 μήνες)

163 εμφανίσεις

Zahra Abbasi

Adel Dokhanchi


Introduction


Problem description:



Adaptive cloud based service provisioning


Problem formulation


Formulating the problem as a binary programming
optimization problem


Simulation setup and evaluation



Virtualized network/Cloud computing


The detail of infrastructure is hidden for service providers and
users


Applications can be hosted in any node in a dynamic fashion


Providing service for mobile users through
clouds


Cloud based services: Infrastructure of the network
and DC are hidden from service provider and users


Service can be hosted in any DC of the cloud


The access point of mobile users changes
over time



Extreme scenarios


Hosting the server in one data center


Hosting the servers in all data center


Adaptive could based service


Dynamically changing the # and location of hosting


Minimizing energy consumption


Maximizing quality of service for mobile users



Cloud computing


New technology


Demand new algorithms/mechanisms for
scheduling, security, accounting



Cloud computing for mobiles



Online or offline computing


Dynamic service migration for mobile users



Dynamic scheduling across data centers


Energy cost model





M=4 data centers


K=10 locations


Each area
a
i

contains
n
i

users


N varies over time


1

2

3

4

10

9

8

7

6

5

1

2

3

4


Mobility of users in each area changes
n
j


d
ij

is the delay from data center
s
i

to area
a
j


M
×
K matrix for delays

1

2

3

4

10

9

8

7

6

5

1

2

3

4

ON

OFF

ON

OFF

d
42

d
43

d
37

d
36

d
35

OFF

OFF

a2

a2

a3

a4

s1

s2

s3

Scheduler

(
onSlots
)

-
Energy cost

-
performance parameters

-
utilization

-
QoS

requirement

-
# of users

X
11

X
31


Computation Energy Cost


Paid to Data Center


Quality of Service Cost


Paid to Mobile User


Delay causes Service Level Violation


Migration Cost


Paid to Virtual Network provider


Imposes Delay

[Kuris et. al.] ICAC 2008

Service Provider

Energy
Cost

$

QoS

Cost

$

Energy
Cost

$


Linear utilization model


u
i
=
nc



Linear power
consumption model



Linear energy cost
model:


z
i
: {0,1}


1
-
>
s
i

is in service


0
-
>
s
i

is NOT in service



ω

Idle power

power



ω

+
α

Maximum power

Ut i lization

0

1


η: paid per user


Migration cost: Setup a new service in a DC
for connected users


Constant migration cost (
β
)


μ
ij
: migrate or not to migrate


Minimize total cost:





Subject to:


All variables are binary.


All users are assigned to a center:



Idle power for non zero utilized servers:


Migration:



Binary programming are generally NP
-
complete

BP=LP for
uni
-
modular constraint matrix (B)

# of
vars
: |A||S|+2|S|

# of
constraints
:
|A|+|S|+|A||S|



Developing a simulator by MATLAB



Solving the problem by GLPK solver
(GLPK+MATLAB)



Verification/evaluation



Uniform mobility pattern


2

1

3

4

10

9

8

7

6

5

1

4

3

2

d
35




Simulation setup improvement


Mobility pattern


Costs


Modeling


Migration modeling


Evaluation



[M.
Bienkowski

et al] “Competitive analysis for service migration in
Vnets
” ACM Virtualized Infrastructure Systems and Architectures,
2010.


K. Kumar et al] “Cloud computing for mobile users: Can off loading
computation save energy?” IEEE Computer, vol. 99, pp. 51

56,
2010.


[M.
Satyanarayanan

et al] “The case for
vm
-
based cloudlets in mobile
computing,” IEEE Pervasive Computing, vol. 8(4), pp. 14

23, 2009.


D.
Kusic

et al] “Power and performance management of virtualized
computing environments via
lookahead

control,” IEEE Cluster
Computing, vol. 12, pp. 1

15, 2009.


[F.
Hermenier

et al] “Entropy: a consolidation manager for clusters,”
ACM Virtual Execution
Environmen
, pp. 41

50 , 2009.