A Survey of Computation Offloading for Mobile Systems

31
A Survey of Computation Offloading for Mobile Systems Karthik Kumar, Jibang Liu Yung-Hsiang Lu , Bharat Bhargava Presented by Zhipeng Yang (Scar) Csci 8980 : Mobile Cloud Computing

Transcript of A Survey of Computation Offloading for Mobile Systems

Page 1: A Survey of Computation Offloading for Mobile Systems

A Survey of Computation Offloading for Mobile Systems

Karthik Kumar, Jibang Liu

Yung-Hsiang Lu , Bharat Bhargava

Presented by

Zhipeng Yang (Scar)

Csci 8980 : Mobile Cloud Computing

Page 2: A Survey of Computation Offloading for Mobile Systems

Introduction1

Technology 2

Analysis3

Future 4

Summary5

Page 3: A Survey of Computation Offloading for Mobile Systems

Demand

Limitation

• Processors

• Storage

• Bandwidths

• Battery

• ……

• Object Recognition

• Real-Time

Translation

• Video Processing

• HD Games

• ……

Page 4: A Survey of Computation Offloading for Mobile Systems

Client/Server --- A thin client ALWAYS migrates computation to a server.

Multiprocessor Systems & Grid Computation --- For load balancing; Migrate for

the SAME environment.

SETI@home --- THOUSANDS of users.

Solution --- Offloading

Computation Offloading is different.

Computation Offloading: PARTLY migrate to (maybe) DIFFERENT

for (maybe) a SINGLE user.

Page 5: A Survey of Computation Offloading for Mobile Systems

• Feasibility (prior to 2000) --- Mainly due to low bandwidths.

• Decision (in early 2000s) --- Algorithms, for better performance & saving energy.

Parameters: Bandwidths, server speeds, memory, server loads, size of

data to be sent,

Solution: Partitioning programs & predicting parametric variations.

• Infrastructures (in late 2000s) --- Virtualization, bandwidths, etc.

• VMs Isolation & Protections Granularities

Development over the last 15 years

Page 6: A Survey of Computation Offloading for Mobile Systems

Classification Factors

• Why to offload

Performance

Save energy

• When to decide offloading

Static

Dynamic

• What mobile systems use offloading

Laptops, PDAs, Robots, Sensors

• Type of applications

Multimedia, gaming, calculators, text editors, predictors

• Infrastructures for offloading

Grid and cloud computing

Page 7: A Survey of Computation Offloading for Mobile Systems

Chroma

(Method)

CuckooMAUI

JIT-Cloudlet

CloneCloud

Comet(Thread)

(VM, Overlay)

Weblet(Component)

Sharing-based(Application)

(Task, Compact Discription)

(Method, Manually)

(Thread, Static)

Page 8: A Survey of Computation Offloading for Mobile Systems

Introduction1

Technology 2

Analysis3

Future 4

Summary5

Page 9: A Survey of Computation Offloading for Mobile Systems

Enabling technologies for computation offloading

(paper counts, obtained from IEEE Xplore)

Page 10: A Survey of Computation Offloading for Mobile Systems

Two significant enablers

1. Wireless networks & mobile agents

• Wireless networks --- WiFi

• Mobile agents

Kotz et al. --- Agent TCL (Tcl, Java, and Scheme)

Wong et al. --- Concordia (Java)

Joseph et al. --- Rover (toolkit,)

--- Migrate computation, platform-independent

2. Virtualization (when?) & cloud computing

• 1960s lost popularity during the 1980s and early 1990s

• Why we use it again?

Under utilization

Security

Operational cost

• Advantages: Multitasking, utilization and flexibility

Page 11: A Survey of Computation Offloading for Mobile Systems

Introduction1

Technology 2

Analysis3

Future 4

Summary5

Page 12: A Survey of Computation Offloading for Mobile Systems

Two main goals of offloading

• Improve performance

• Save energy

Page 13: A Survey of Computation Offloading for Mobile Systems

Performance

sm the speed of the mobile system

ss the speed of the server

w the amount of computation that can’t be migrated

di the size of data need to be sent

B the bandwidth

How does the inequality hold?

Large w & small di

Page 14: A Survey of Computation Offloading for Mobile Systems
Page 15: A Survey of Computation Offloading for Mobile Systems

Energy

pm the power on the mobile system

pm the power required to send data form mobile system over the network

pm the power required to wait for the result

sm the speed of the mobile system

ss the speed of the server

w the amount of computation that can’t be migrated

di the size of data need to be sent

B the bandwidth

How does the inequality hold?

Page 16: A Survey of Computation Offloading for Mobile Systems
Page 17: A Survey of Computation Offloading for Mobile Systems

Classification criteria here

• When to decide offloading & How are tasks identified for offloading

Static

Dynamic

• How are tasks identified for offloading

• Type of applications

Multimedia, gaming, calculators, text editors, predictors

• What mobile systems use offloading

Laptops, PDAs, Robots, Sensors

Page 18: A Survey of Computation Offloading for Mobile Systems
Page 19: A Survey of Computation Offloading for Mobile Systems
Page 20: A Survey of Computation Offloading for Mobile Systems

• Static, low overhead & need accurate prediction

Prediction algorithms: probabilistic prediction, history-based

prediction, fuzzy control

• Dynamic, flexible & high overhead with monitors

Prediction (Bayesian scheme) + monitors

Static &Dynamic

Page 21: A Survey of Computation Offloading for Mobile Systems

Identification

• Various algorithms for partition.

• E.g., Communication should be taken in to account.

Size of a vertex indicates the amount of computation

Width of an edge indicates the amount of data to be sent

Page 22: A Survey of Computation Offloading for Mobile Systems

Applications & Systems

Page 23: A Survey of Computation Offloading for Mobile Systems

Applications & Systems

Page 24: A Survey of Computation Offloading for Mobile Systems

Infrastructures

• Inter-operability, different networks, systems and capabilities

• Mobility and fault tolerance, handle failures

• Privacy & Security, for user’s confidential data

• Context awareness, user’s location and context

Page 25: A Survey of Computation Offloading for Mobile Systems
Page 26: A Survey of Computation Offloading for Mobile Systems

Introduction1

Technology 2

Analysis3

Future 4

Summary5

Page 27: A Survey of Computation Offloading for Mobile Systems

Two important trends

• Sensor deployment, not for computation but for data collection

and protection

• Growth in smartphones

Page 28: A Survey of Computation Offloading for Mobile Systems

• More requirements on mobile system.

Recognition, real-time translation, data management etc.

• Massive growth in mobile data in both types and volumes

Images, audios, videos etc.

• The capacities of the devices will be unlikely to grow at the

same pace.

Processors, storage, bandwidths, battery etc.

Page 29: A Survey of Computation Offloading for Mobile Systems

Introduction1

Technology 2

Analysis3

Future 4

Summary5

Page 30: A Survey of Computation Offloading for Mobile Systems

Summary

• What was done in this paper.

Background in details

Enabler technology

Simplified formula with extended example

Statistics from a lot of papers

Simple analysis about future trend

• Some comments.

Most of the issues above are simplified for readers

Easy to understand but not accurate, e.g. formula

More deep comparisons are needed for algorithms

Data sources are limited

Page 31: A Survey of Computation Offloading for Mobile Systems

Thank you