• No results found

Non-Cooperative Computation Offloading in Mobile Cloud Computing

N/A
N/A
Protected

Academic year: 2021

Share "Non-Cooperative Computation Offloading in Mobile Cloud Computing"

Copied!
19
0
0

Loading.... (view fulltext now)

Full text

(1)

Non-Cooperative Computation

Offloading in Mobile Cloud Computing

Valeria Cardellini

University of Roma Tor Vergata, Italy

Joint work with: V. De Nitto Personè, V. Di Valerio, F. Facchinei, V. Grassi, F. Lo Presti, V. Piccialli

Joint CLEEN and ACROSS Workshop on Cloud Technology and Energy Efficiency in Mobile Communications

(2)

Computation offloading

n 

Mobile devices have

limited

computing and

processing

capabilities

¨ Battery life!

n 

“Old” idea:

offload computation

to some

surrogate server (cyber foraging)

n 

Mobile-cloud convergence

n 

Increasing number of latency-sensitive,

compute-intensive and memory-intensive

applications on mobile devices

¨ Wearable cognitive assistance, edge analytics in IoT,

automotive environments, …

(3)

Cloud-­‐based  augmenta0on  of  mobile  

devices

 

n 

Resource-­‐limited  mobile  devices  can  overcome  their  

limita0ons  by  offloading  opera0ons  to  more  

powerful  servers  

¨ Either  proximate,  i.e.  single  hop  (cloudlet)  or  mul0-­‐hop  

(cloud  data  center)  

Mobile devices Cloudlet Distant cloud servers on Internet Wireless AP

3 Non-Cooperative Computation Offloading in Mobile Cloud Computing @EUCNC'15 Workshop 1, June 29 2015

(4)

Cloudlet  

n 

Self-­‐managed  data  center  with  soC  state  and  located  

at  the  network  edge  

¨ Consistent  with  Fog  compu0ng  (“Fog  is  the  cloud  close  to  

the  ground”)  and  micro  cloud  concepts  

n 

Example  of  cloudlet  infrastructure  and  soCware  

 

Nokia  LTE  base  sta0on  with  cloudlet-­‐like   hos0ng  infrastructure  

Cloudlet-­‐specific  extensions  to   OpenStack  (by  Satya  at  CMU)  

(5)

Three-­‐0er  architecture  

n 

Hybrid  three-­‐0er

 architecture  for  mobile-­‐cloud  

convergence  

¨ Local  0er:  mobile  nodes  

¨ Middle  0er:  energy-­‐unlimited  but  resource-­‐limited  nearby  

cloudlets  within  one  wireless  hop  of  mobile  devices  

¨ Remote  0er:  resourceful  distant  cloud  servers    

n 

Where  to  offload?

 

¨ How  to  get  the  benefits  of  external  servers  for  offloading  

while  limi0ng  latency  issues  of  distant  servers  and   considering  energy  constraints  of  mobile  devices  

(6)

Our  proposal  

n 

Non-­‐coopera0ve

 game  theore0c

 approach  to  

analyze  and  manage  the  computa0on  offloading  

strategy    

¨ Unmanaged  usage  scenario:  no  central  authority  

¨ Mul0ple  non-­‐coopera0ve  mobile  devices  share  the  

cloudlet  limited  compu0ng  resources    

¨ Mobile  devices  can  send  their  computa0ons  to  any  of  the  

three  0ers  

V.  Cardellini,  V.  De  NiTo  Personè,  V.  Di  Valerio,  F.  Facchinei,  V.  Grassi,  F.  Lo  Pres0,   V.  Picciali,  “A  game-­‐theore0c  approach  to  computa0on  offloading  in  mobile  cloud   compu0ng”,  Mathema5cal  Programming,  available  online,  Apr.  2015.  

(7)

Our  goal  

n 

Determine  

whether  and  where  to  offload  

a  task  

based  on  the  impact  this  has  on  the  user  experience,  

expressed  through  

QoS  measures    

(8)

System  model  

n 

Standard  queueing  

theory  for  system  model  

¨ To  capture  the  effects  of  

computa0on  offloading   (i.e.,  task  conten0on  on   shared  resources)  on   users’  perceived  

performance  

¨ Focus  on  task                

response  0me  

(9)

Problem  formula0on  

n 

To  determine  the  

op0mal  strategy,  each  

user  needs  to  solve:  

n 

Problem  formula0on:  

generalized  Nash  equilibrium  

problem

 (GNEP)  

¨ Both  the  players  objec0ve  func0ons  and  the  strategy  sets  

depend  on  the  other  players  strategies  

(10)

GNEP  proper0es  and  distributed  algorithm  

n 

Game  can  be  solved  by  finding  a  solu0on  to  a  

suitable  

Varia0onal  Inequality  

¨ We  show  equilibrium  existence  and  monotonicity  

proper0es  of  the  GNEP  

n 

Distributed  algorithm

 for  compu0ng  an  

equilibrium  strategy  for  each  mobile  user  

n 

Its  design  required  us:    

¨ To  analyze  in-­‐depth  the  underlying  equilibrium  

problem  

¨ To  apply  some  very  recent  algorithmic  developments   n 

Appealing  for  real-­‐world  implementa0on  

(11)

Distributed  algorithm

n 

Synchronous  and  itera0ve  

(12)

Numerical  results  

Non-Cooperative Computation Offloading in Mobile Cloud Computing @EUCNC'15 Workshop 1, June 29 2015 12

cloudlet  

mobile  

cloudlet  

cloud  

(13)

Numerical  results  

Non-Cooperative Computation Offloading in Mobile Cloud Computing @EUCNC'15 Workshop 1, June 29 2015 13

n 

Price of anarchy

cloudlet  

cloud   cloudlet  

(14)

Prototype  architecture  

Non-Cooperative Computation Offloading in Mobile Cloud Computing @EUCNC'15 Workshop 1, June 29 2015 14

n 

Working  prototype  

based  on  offloading  middleware  

(15)

Prototype  addi0onal  features  

n 

Adapta0on  capabili0es  based  on  MAPE  cycle    

n 

Discovery  and  deployment  of  offloading  service    

(16)

Prototype  results:  face  recogni0on  

Non-Cooperative Computation Offloading in Mobile Cloud Computing @EUCNC'15 Workshop 1, June 29 2015 16

(17)

Conclusions  

n 

Non-­‐coopera0ve  usage  scenario  for  computa0on  

offloading  in  a  three-­‐0er  architecture    

¨ Users  try  to  take  advantage  selfishly  of  the  available  

cloudlet  resources  

¨ Solu0on  with  solid  theore0cal  founda0on  and  appealing  

for  real-­‐world  implementa0on  

(18)

Future  work  

n 

Possible  extension  of  our  proposal  

¨ Remove  limita0on  on  user  generated  load  (cannot  exceed  

the  mobile  device  capacity)  

¨ Consider  mul0ple  class  model  of  tasks  launched  by  users  

¨ Include  monetary  cost   n 

More  general  challenges  

¨ Use  cloudlets  to  improve  reliability  

¨ Consider  mul0ple  (possibly  coopera0ve)  cloudlets  

(19)

That’s  All  Folks!  

19

Thank  you  for  the  aTen0on  

 

cardellini@ing.uniroma2.it  

hTp://www.ce.uniroma2.it/~valeria  

References

Related documents

• Distinguishing between hypertensive emergency (associated with acute target organ damage) and urgency (no target organ damage) is crucial to appropriate management.. •

Finally, the two theoretical arenas of collaborative planning and the just city are reviewed in order to evaluate current planning practices pertaining to refugees and asylum

DIRECCIÓN GENERAL DE ADMINISTRACIÓN, EVALUACIÓN Y CONTROL

To address this, the JISC Usage Statistics Review Project aims to formulate a fundamental scheme for recording usage data and to propose a standard for its aggregation to

• Experience in developing naturalist education programs, California native plant propagation and community based restorations within the non-profit sector.. • Cris joined the

Table 9 presents the estimates of the labor market returns to speaking a foreign language using college and high school graduation requirements as instruments for the ability to speak

N • CMOS output • LCD output • CMOS input • Hysteresis input • Automotive input Q • CMOS output • LCD output • Hysteresis input • Automotive input R • CMOS output •

interventions, maternal and neonatal outcomes for all women who used a birthing pool during labour who either had a waterbirth or left the pool and had a landbirth, and for the