The Case for VM-based Cloudlets in Mobile Computing

THE CASE FOR VM-BASED CLOUDLETS IN

MOBILE COMPUTING

Source: IEEE Pervasive Computing, Vol. 8, Issue.4, Oct.2009, pp. 14 – 23Author: Satyanarayanan, M. , Bahl, P. , Caceres, R. , Davies, N.

Adviser: Chia-Nian ShyiSpeaker: 趙家寬Date: 2010/11/2 

1

Introduction• We have developed the core concepts,

techniques and mechanisms for mobile computing.

• Mobile computing seamlessly augments the cognitive abilities of users.Speech recognitionNatural language processingComputer vision and graphicsMachine learningAugmented realityPlanning and decision-making

2

Introduction• Exploit virtual Machine (VM) technology

to rapidly instantiate customized service software on a nearby cloudlet, then uses that service over a wireless LAN.

• Cloudlet also simplifies the challenge of meeting bandwidth demand of multiple users, such as HD video and high-resolution images.

3

Resource-Poor Mobile Hardware Mobile hardware is necessarily

resource-poor relative to static client and server hardware.

Improving size, weight and battery life are higher priorities than enhancing compute power.

Resource poverty is a major obstacle for many applications with the potential to seamlessly augment human cognition.

4

The Limits of Cloud Computing Cloud computing - solution to the

resource poverty of mobile devices. Long WAN latencies are a fundamental

obstacle. WAN delays in the critical path of user

interaction can hurt usability.

5

The Limits of Cloud Computing Lagar-Cavilla et al shows latency can

negatively impact interactive response in spite of adequate bandwidth.

6

How Cloudlets Can Help Resource poverty of a mobile device

can be addressed by using a nearby resource-rich cloudlet.

The end-to-end response time of applications executing in the cloudlet needs to be fast (few milliseconds) and predictable.

7

How Cloudlets Can Help Cloudlets are decentralized and widely-

dispersed Internet infrastructure.

8

How Cloudlets Can Help A cloudlet only contains soft state such

as cache copies of data or code that is available elsewhere.

9

Transient Cloudlet Customization

A key challenge is to simplify cloudlet management.

Solution is transient customization of cloudlet infrastructure using hardware virtual machine (VM) technology.

Pre-use customization and Post-use cleanup.

10

Transient Cloudlet Customization

Two different approaches to delivering VM state to infrastructure.

VM migration approach Already-executing VM is first suspended,

processor, disk and memory state are then transferred, VM execution is resumed at the destination.

11

Transient Cloudlet Customization

The other approach is called dynamic VM synthesis.

12

Transient Cloudlet Customization

Dynamic VM synthesis differs cloud in two key ways. First its performance is determined solely by local resources:Bandwidth to cloudlet Compute power of the cloudlet

Second, WAN failures do not affect synthesis.

13

Dynamic VM Synthesis A proof-of-concept prototype called

Kimberley. The controller of the transient binding

between mobile device and cloudlet called Kimberley Control Manager (KCM).

14

Dynamic VM Synthesis The first step in the binding sequence is

the establishment of a secure TCP tunnel using SSL between KCM on a device and a cloudlet.

After successful authentication, the cloudlet KCM executes a command that fetches the VM overlay from the mobile device.

15

Dynamic VM Synthesis Decrypts and decompresses it, and

applies the overlay to the base VM.

The VM is then launched, and is ready to provide services to the mobile device.

16

Dynamic VM Synthesis For use in cloudlets, rapid VM synthesis

is important. Cloudlet handoffs should be as rapid,

invisible and seamless.

17

Dynamic VM Synthesis Significant improvement is needed for

real-world deployment. The major contributors to VM synthesis

time are (a) overlay transmission and (b) decompressing and applying the overlay on the cloudlet.

18

Dynamic VM Synthesis To reduce decompression and overlay

application times, can exploit parallelism.

Partitioning the VM image into four parts will allow a four-core cloudlet to synthesize the parts.

19

Conclusion Resource poverty is a fundamental

constraint. A vision of mobile computing that breaks

free of this fundamental constraint. Mobile users seamlessly utilize nearby

computers to obtain the resource benefits of cloud computing without incurring WAN delays and jitter.

20