Cloudlet-Based Cyber-Foraging in Resource-Constrained Environments

© 2013 Carnegie Mellon University

Cloudlet-Based Cyber-Foraging in Resource-Constrained Environments

VU University Amsterdam September 26, 2013 Grace A. Lewis ([email protected])

2

Cloudlet-Based Cyber-Foraging in Resource-Constrained Environments © 2013 Carnegie Mellon University

Motivation First responders and others operating in crisis and hostile environments increasingly are making use of handheld devices to help with tasks such as speech and image recognition, natural language processing, decision-making and mission planning Challenges for mobile devices in resource-constrained edge environments •  Mobile devices offer less computational power than

conventional desktop or server computers •  Computation-intensive tasks take a heavy toll on

battery power •  Networks are often unreliable and bandwidth is

limited and inconsistent

3


Characteristics of Resource-Constrained Environments Wimpy edge nodes •  Limited resources (CPU, battery and memory) on mobile nodes

Limited or no end-to-end network connectivity •  Implicit assumption of WAN connectivity is not always valid

High cognitive load •  Application latency and fidelity become important in situations of high stress

Bounded elasticity •  Upper bound on number of consumers known in advance

Dynamic environment •  Static deployment topologies cannot be assumed •  Survivability is essential

4


Mobile Cloud Computing

Mobile Cloud Computing refers to the combination of mobile devices and cloud computing in which cloud resources perform computing-intensive tasks and store massive amounts of data Work in this field is currently divided into three areas 1.  Extending access to cloud services to mobile devices 2.  Enabling mobile devices to work collaboratively as cloud resource

providers 3.  Augmenting execution of mobile applications using cloud resources

Cyber-foraging is part of the last area

5


Cyber-Foraging

Cyber-foraging* is the leverage of external resource-rich surrogates to augment the capabilities of resource-limited devices •  Increase computational capability •  Conserve battery on a mobile device •  Access remote or shared data •  Provide better application fidelity

Two main forms of cyber-foraging •  Code/Computation Offload – Offload of expensive computation in order to extend battery life and

increase computational capability •  Data Staging –  Improve data transfers between mobile computers and the cloud by

temporarily staging data in transit

* Satyanarayanan, Mahadev (2001). "Pervasive Computing: Vision and Challenges". IEEE Personal Communications (IEEE)

6


Challenges for Cyber-Foraging in Resource-Constrained Environments

Most cyber-foraging solutions rely on •  conventional Internet for connectivity to the cloud •  strategies that tightly couple mobile clients with servers at deployment time

These solutions are not appropriate for resource-constrained environments because they do not address the challenge of unreliable networks and dynamic environments

7


Cyber-Foraging Questions

Where to offload? •  Proximate (single-hop) or remote (multi-hop) servers?

When to offload? •  With optimization in mind, when does it make sense to offload?

What to offload? •  What is the granularity of the computation (or data) that is offloaded? •  What is the payload?

8


Cyber-Foraging: Where to Offload?

Mobile Enterprise Cloud

(Data Center)

Mobile Surrogate Enterprise Cloud (Data Center)

Synchronous network I/O, Multi-hop network

Synchronous Network I/O

Asynchronous Network I/O

tcloud

tsurrogate

Single-hop network

tcloud

Assuming that tsurrogate is less than tcloud, nearby surrogates are a better option from an energy consumption and latency perspective

9


Cyber-Foraging: When to Offload?

Never offload

It Depends: Size of this area depends on

optimization algorithm

Always Offload

Offloading is beneficial when large amounts of computation (C) are needed with relatively small amounts of communication (D)*

Goal of Optimization •  Maximize a utility

function •  Typically energy,

execution time, fidelity or results

Computation (C)

Com

mun

icat

ion

(D)

* Based on Karthik Kumar and Yung-Hsiang Lu. Cloud computing for mobile users. Computer, 99, 2010.

10


Cyber-Foraging: What to Offload?

Another aspect of what to offload is the actual payload •  Computation is transferred from the mobile device to the surrogate at runtime,

or •  Code already exists on the surrogate and is remotely invoked, in which case

the payload is simply the input parameters

Physical Machine

OS

Infrastructure

VM Image (VM Manager, e.g., KVM, OpenStack)

Process/Application/Service (Application Server)

Method/Thread (Language-Specific Virtual Machine, e.g., JVM, CLR)

Surrogate

Mobile Code (Runtime Environment)

One aspect is granularity

11


Cloudlet-Based Cyber-Foraging Discoverable, generic, stateless servers located in single-hop proximity of mobile devices Can operate in disconnected mode — communication with the central core is only needed for provisioning Virtual-machine based to promote flexibility and mobility Applications statically partitioned into client and server •  Client runs on mobile device •  Server runs on cloudlet

Enterprise Cloud

Cloudlet Cloudlet Cloudlet

Single-‐Hop Network

Multi-‐ or Single-‐Hop Network

Cloudlet

12


Reference Architecture for Cloudlet-Based Cyber-Foraging

Cloudlet HostMobile Client

Legend

System Boundary

Cloudlet-‐Ready

Client App 1

Custom Runtime Component

Cloudlet Client

File Read/WriteCall

File

VM Manager

Guest VM 1

Server 1

Cloudlet Server

Discovery Service

Cloudlet Metadata

Broadcast

3rd Party Runtime

Component

Server Offload Code + Metadata

Cloudlet-‐Ready

Client App 2

Cloudlet-‐Ready

Client App n...

...

Guest VM 2

Server 2

Guest VM n

Server n

...

Cloudlet Metadata

Client App 1

Metadata

Client App n

Metadata

Client App 2

Metadata

Server 1 Offload Code

Server 2 Offload Code

Server n Offload Code

13


Original Proposal for Cloudlet Provisioning: VM Synthesis

Application overlay is transmitted to the cloudlet at runtime Cloudlet applies application overlay to the Base VM and produces a Complete VM

Start Base VM

Base VM Disk Image

Install Application

Complete Memory

Snapshot

Suspend VM

Save Disk and

Memory Snapshots

Complete Disk

Snapshot

Obtain Base VM from

Central Core

Start Application

Base Memory

Snapshot

Calculate Diff Between Complete Snapshots and

Base Snapshots

VM Disk Image

Overlay

Memory Snapshot Overlay

Application Overlay

Base Disk

Snapshot

14


VM Synthesis Process 1. Mobile device

discovers cloudlet 2. Mobile device sends

application overlay (always offload strategy)

3. Cloudlet applies overlay to base VM

4. Cloudlet starts the constructed VM and informs client that the application server is ready for execution


Legend

System Boundary

Cloudlet-Ready Client App 1


Cloudlet Client

Application Overlay n

File Read/WriteCall

File

VM Manager

Guest VM 1

Server 1

Cloudlet Server

Discovery Service Cloudlet Server IP Address/Port

Broadcast

Base VM Image

3rd Party Runtime

Component

Application Overlay 2

Application Overlay 1

Application Overlay

Cloudlet-Ready Client App 2

Cloudlet-Ready Client App n

...

...

Guest VM 2

Server 2

Guest VM n

Server n

...

15


Problems with VM Synthesis in Resource-Constrained Environments VM overlays are very large •  Large payloads incur large energy costs •  Limited bandwidth can make application ready time unreasonable

Payload NULL OBJECT SPEECH FACE

Linux (MB) Windows (MB) Linux (MB) Linux

(MB) Windows

(MB) Windows

(MB)

Compressed Disk Overlay 0.1 0.4 42.8 104.8 113.7 33.5

Compressed Disk + Memory Overlays

21.2 4.2 144.5 226.8 425.7 141.5

Question: Is it possible to maintain the flexibility and mobility of a VM-based strategy but reduce payload size?

16


Cloudlet Provisioning using Application Virtualization

Emulates OS functionality for an application Runtime component intercepts all system calls from an application and redirects these to resources inside the virtualized application Application itself is unmodified and unaware that it is interacting with virtual OS services Sample Tools •  CDE (short for Code, Data and Environment) – Application virtualizer for Linux – Virtualizes applications by monitoring their execution

•  Cameyo – Application virtualizer for Windows – Virtualized applications by monitoring the installation process

17


Application Virtualization Process

Mobile Device

Application 1 Package

Cloudlet Host

VM Hypervisor

Guest VM – OS 1

Cloudlet Server

Application 1 Server

Application i Server

Guest VM – OS n

Discovery

Application 1Client

Application 1 Metadata

HTTP Server

Cloudlet Server

Application j Server

Application m Server

Discovery

HTTP Server

Cloudlet Client

Discovery

Application iClient

Application jClient

Application mClient

Application iPackage

Application i Metadata

Application j Package

Application j Metadata

Application m Package

Application m Metadata

Appl ication Package

Appl ication Metadata

Appl ication Package

Appl ication Metadata

IP Address : PortCloudlet Properties

Legend

System Boundary Custom RuntimeComponent

3rd Party RuntimeComponent

Fi le Blocking Call Non-‐Blocking Call Fi le Read HTTP Multicast

1. Mobile device discovers cloudlet

2. Mobile device sends application package and metadata

3. Cloudlet deploys application package inside matching guest VM

4. Cloudlet informs client that the application server is ready for execution

18


VM Synthesis vs. Application Virtualization

Cyber-Foraging Strategy

Transmitted Package

NULL OBJECT SPEECH FACE

Linux (MB) Windows (MB) Linux (MB) Linux

(MB) Windows

(MB) Windows

(MB)

Application Virtualization

Compressed Virtualized Application

1.1 0.9 28.5 67.7 65.4 13.1

VM Synthesis

Compressed Disk Overlay 0.1 0.4 42.8 104.8 113.7 33.5

Compressed Disk + Memory Overlays 21.2 4.2 144.5 226.8 425.7 141.5

Experiments show that energy consumption is proportional to file size Even though application virtualization creates payloads of smaller size, it is still large for resource-constrained environments Exploring two alternative options •  Cloudlet Push •  On-Demand VM Provisioning

19


Cloudlet Push

Opposite approach to traditional cyber-foraging and cloudlet provisioning •  Pushes client application from the cloudlet to the mobile device

20


Cloudlet Push (PROTOTYPE)

Legend

System Boundary

Client App 1


Cloudlet Client

File Read/WriteCall

File

VM Manager

Guest VM 1

Server 1

Cloudlet Server

Discovery Service Cloudlet Metadata

Broadcast

3rd Party Runtime

Component

Client App

Client App 2

Client App n

...

Guest VM 2

Guest VM n

Server n

...

Cloudlet Metadata

Client Metadata

Client App

Server 2

Client Metadata

Cloudlet HostCloudlet HostMobile ClientMobile Client

1. Mobile device discovers cloudlet (and available applications)

2. Mobile device sends client metadata and selected app

3. Cloudlet sends client app

4. Mobile device installs and starts client app

21


On-Demand Provisioning

Leverages commercial cloud provisioning tools such as Puppet or Vagrant to “compose” VMs on demand to match the requirements of the mobile application client •  What is transferred from the mobile device to the cloudlet is a provisioning

script Composition could be •  Instantiating a pre-configured VM from a VM image repository that contains

the full server component •  Instantiating a pre-configured base VM and adding components •  Putting together a VM from a set of basic components

22


On-Demand VM Provisioning (WORK IN PROGRESS)

1.  Mobile device discovers cloudlet

2.  Mobile device sends provisioning script to cloudlet

3.  Cloudlet runs VM provisioning script and starts VM

4.  Cloudlet informs client that the application server is ready for execution


Legend

System Boundary

Cloudlet-‐Ready

Client App 1


Cloudlet Client

File Read/WriteCall

File

VM Manager

Guest VM 1

Server 1

Cloudlet Server

Discovery Service

Cloudlet Metadata

Broadcast

3rd Party Runtime

Component

Provisioning Script

Cloudlet-‐Ready

Client App 2

Cloudlet-‐Ready

Client App n

...

...

Guest VM 2

Server 2

Guest VM n

Server n

...

Cloudlet Metadata

Server 1 Provisioning

Script

Server n Provisioning

Script

Server 2 Provisioning

Script

VM Image Repository

Database

23


Comparison of Cloudlet Provisioning Strategies VM Synthesis Application

Virtualization Cloudlet Push On-Demand VM

Provisioning

Cloudlet Contents (plus Cloudlet Server)

Exact Base VM(s) VM(s) compatible with client app(s)

•  Paired VM (server code)

• Client app(s)

•  VM provisioning software

•  Server components

Mobile Device Contents

• Cloudlet client •  Application

overlay • Client app

• Cloudlet client •  Virtualized server

code • Client app

• Cloudlet client • Cloudlet client •  VM provisioning

script(s) • Client app

Advantages Guarantees replica of pre-configured surrogate

Portability across OS distribution boundaries

Supports most client nodes at runtime

Small data transfer

Constraints Exact Base VM — limits distribution and patches

All dependencies must be captured at packaging time

Cloudlet has a client app version matching mobile client OS version

Cloudlet has all required VM components

Distribution to Mobile Devices

Deployment Deployment Runtime Deployment

Runtime Transfer Cost

Depends on overlay size

Medium Low Very low

24


Current Work: Architecture Strategies for Cyber-Foraging in Resource-Constrained Environments Research Question What are the strategies/patterns/tactics — expressed as architecture decisions and promoted quality attributes — that would provide architecture alternatives for cyber-foraging that adapt to unreliable networks and dynamic environments while conserving battery power

25


Related Work 1

Work in cyber-foraging dates back to 2001 when the term was coined •  Most proposed solutions for cyber-foraging come from electrical and

computer engineering with very little emphasis on software engineering or software architecture

•  Most solutions favor one quality attribute (typically battery consumption or latency) with very little consideration of architecture tradeoffs

•  Many solutions assume that code can execute on both the mobile device and the surrogate, which is highly unlikely for the complex algorithms used by edge applications

•  Many solution tightly couple mobile clients with surrogates at deployment time

26


Related Work 2

Representative Cyber-Foraging Solutions

Quality Attributes Basic Tradeoff Space Extended Tradeoff Space

Ener

gy

Exec

utio

n Ti

me

Fide

lity

of

Res

ults

Net

wor

k C

apab

ilitie

s

Surv

ivab

ility

Res

ilien

cy

Secu

rity

Flex

ibili

ty

Spectra [Flinn 2002] X X X

Chroma [Balan 2003] X X X

Surrogates [Goyal 2004] X X

Slingshot [Su 2005] X X X X

Cloudlets [Satyanarayanan 2009] X X

MAUI [Cuervo 2010] X X X X

Scavenger [Kristensen 2010] X X

CloneCloud [Chun 2011] X X X

Odessa [Ra 2011] X X

27


Goal

Identify architecture strategies that address a much broader range of quality attributes that are critical in resource-constrained environments, e.g. •  Survivability — fitness/readiness, awareness, anticipation, speed

(responsiveness), agility (e.g., flexibility/ maneuver), and evolvability* •  Resiliency — robustness (e.g., redundancy), diversity, active defense, rapid

reconstitution*

* Source: Air Force Cyber Vision 2025. July 2012.

28


Contact Information

Grace A. Lewis Advanced Mobile Systems (AMS) Initiative Software Engineering Institute 4500 Fifth Avenue Pittsburgh, PA 15213-2612 USA Phone: +1 412-268-5851 Email: [email protected] WWW: http://www.sei.cmu.edu/staff/glewis/

29


Selected Cyber-Foraging References [Balan 2003] Rajesh K. Balan, Mahadev Satyanarayanan, So Young Park, and Tadashi Okoshi. Tactics-based remote execution for mobile computing. In Proceedings of the 1st International Conference on Mobile Systems, Applications and Services, pages 273–286, San Francisco, CA, May 2003. [Chun 2011] Byung-Gon Chun, Sunghwan Ihm, Petros Maniatis, Mayur Naik, and Ashwin Patti. CloneCloud: Elastic execution between mobile device and cloud. In Proceedings of the European Conference on Computer Systems, Salzburg, Austria, April 2011 [Cuervo 2010] Eduardo Cuervo, Aruna Balasubramanian, Dae ki Cho, Alec Wolman, Stefan Saroiu, Ranveer Chandra, and Paramvir Bahl. MAUI: Making smartphones last longer with code offload. In Proceedings of the 8th International Conference on Mobile Systems, Applications and Services, pages 49–62, San Francisco, CA, June 2010. [Flinn 2002] Jason Flinn, SoYoung Park, and Mahadev Satyanarayanan. Balancing Performance, Energy, and Quality in Pervasive Computing. In Proceedings of the 22nd International Conference on. Distributed Computing Systems, Vienna, Austria, July 2002. [Flinn 2012] Jason Flinn. Cyber Foraging: Bridging Mobile and Cloud Computing. Synthesis Lectures on Mobile and Pervasive Computing. Vol. 7, No. 2, pp. 1-103. 2012. [Goyal 2004] Sachin Goyal and John Carter. A lightweight secure cyber foraging infrastructure for resource-constrained devices. In Proceedings of the 6th IEEE Workshop on Mobile Computing Systems and Applications, Lake Windermere, England, December 2004. [Kristensen 2010] Mads Daro Kristensen. Scavenger: Transparent deployment of efficient cyber foraging applications. In IEEE International Conference on Pervasive Computing and Communications (PerCom), pages 217–226, San Diego, CA, March 2010. [Ra 2011] Moo-Ryong Ra, Anmol Sheth, Lily Mummert, Padmanabhan Pillai, DavidWetherall, and Ramesh Govidan. Odessa: Enabling interactive perception applications on mobile devices. In Proceedings of the 9th International Conference on Mobile Systems, Applications and Services, Washington, DC, June 2011. [Satyanarayanan 2009] Mahadev Satyanarayanan, Paramvir Bahl, Ramón Cáceres, and Nigel Davies. The case for VM-based cloudlets in mobile computing. IEEE Pervasive Computing, 8(4):14–23,October–December 2009. [Su 2005] Ya-Yunn Su and Jason Flinn. Slingshot: Deploying stateful services in wireless hotspots. In Proceedings of the 3rd International Conference on Mobile Systems, Applications and Services, pages 79–92, Seattle, WA, June 2005.

30


Our Publications Refereed •  Lewis, Grace; Simanta, Soumya; Novakouski, Marc; Cahill, Gene; Boleng, Jeff; Morris,

Edwin; and Root, James. Architecture Patterns for Mobile Systems in Resource-Constrained Environments. 32nd Annual Military Communications Conference (MilCom 2013). Novermber, 2013.

•  Simanta, Soumya; Lewis, Grace; Morris, Ed; Ha, Kiryong; and Satyanarayanan, Mahadev. A Reference Architecture for Mobile Code Offload in Hostile Environments. Proceedings of the 4th International Conference on Mobile Computing, Applications and Services (MobiCASE 2012). October 2012.

•  Simanta, Soumya; Lewis, Grace; Morris, Ed; Ha, Kiryong; and Satyanarayanan, Mahadev. A Reference Architecture for Mobile Code Offload in Hostile Environments. Proceedings of the Joint 10th Working IEEE/IFIP Conference on Software Architecture & 6th European Conference on Software Architecture (WICSA/ECSA 2012). August 2012.

Other •  S. Simanta, G. Lewis, E. Morris, K. Ha, and M. Satyanarayanan, "Cloud Computing at

the Tactical Edge," Software Engineering Institute, Carnegie Mellon University, Pittsburgh, Pennsylvania, Technical Note CMU/SEI-2012-TN-015, 2012. http://www.sei.cmu.edu/library/abstracts/reports/12tn015.cfm

31


Copyright 2013 Carnegie Mellon University

This material is based upon work funded and supported by the Department of Defense under Contract No. FA8721-05-C-0003 with Carnegie Mellon University for the operation of the Software Engineering Institute, a federally funded research and development center.

Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the United States Department of Defense.

NO WARRANTY. THIS CARNEGIE MELLON UNIVERSITY AND SOFTWARE ENGINEERING INSTITUTE MATERIAL IS FURNISHED ON AN “AS-IS” BASIS. CARNEGIE MELLON UNIVERSITY MAKES NO WARRANTIES OF ANY KIND, EITHER EXPRESSED OR IMPLIED, AS TO ANY MATTER INCLUDING, BUT NOT LIMITED TO, WARRANTY OF FITNESS FOR PURPOSE OR MERCHANTABILITY, EXCLUSIVITY, OR RESULTS OBTAINED FROM USE OF THE MATERIAL. CARNEGIE MELLON UNIVERSITY DOES NOT MAKE ANY WARRANTY OF ANY KIND WITH RESPECT TO FREEDOM FROM PATENT, TRADEMARK, OR COPYRIGHT INFRINGEMENT.

This material has been approved for public release and unlimited distribution. This material may be reproduced in its entirety, without modification, and freely distributed in written or electronic form without requesting formal permission. Permission is required for any other use. Requests for permission should be directed to the Software Engineering Institute at [email protected].

DM-0000299

Technology

Cloudlet-Based Cyber-Foraging in Resource-Constrained Environments