Environment mobility moving the desktop around

1
Environment mobility - moving the desktop around

• Håvard D. Johansen
• joint work with
• Dag Johansen and
• Robbert van Renesse

WAIF project University of Tromsø
2
Outline

• Context and problem.
• Software mobility.
• Lessons so far.

3
Software Mobility Emerging

• Observation 1 All kind of mobility emerging.
• Observation 2 Software not really built for
  mobility.
• ? Conjecture Software mobility should be a
  first order design principle.
• ? Research Goal Devise (potentially orthogonal)
  structures supporting software mobility.

4
WAIF (Wide Area Information Filtering)

• Problem How to structure the next generation
  web.
• Approaches
• 1. Use the network as a personal computer
  (computers/user gtgt 1 )
• ? move personalized computational tasks around.
• 2. Mobile users in a pervasive computing
  environment
• ? design for mobility.
• 3. Proactive computing combined with high
  precision
• ? humans not in the loop, but above the loop.
• International cooperation University of Tromsø,
  Cornell University, and UC San Diego.

5
Big Picture PONS

• Locate services and install your personal
  distributed system as an overlay network.

6
Support for mobile (PONS) users

• Move user environment transparently.

7
Outline

• Context and problem.
• Software mobility.
• Lessons so far.

8
Simple (same) Problem The desktop
Home
9
Challenge move the environment
Home
Work
10
Approach 1 Mobile Machine

• Physical movement of computing platform.

Home
Work
11
Approach 2 Remote Access

• Rewire application input and output.
• Problem Network bandwidth/latency

Desktop
Desktop
thin client
presentation
interaction
Home
Work
12
Our approach Migrate Applications

• End-to-end user experience.
• Migrate just enough state (what is enough state?)

M A S
M A S
Home
Work
13
we are not alone

• Internet Worms (Xerox 82)
• Process migration 85-90 (V, Amoeba, Eden,
  Mach).
• Condor a like checkpoint/restart (88).
• Mobile code and mobile agents (TACOMA 93, 100s
  of systems).
• Mobile Virtual Machines (Stanford 02).
• Aura (Intel and CMU, 02)
• .....

14
Task Migration

• Abstract description
• What the user wants to do
• Audio playback, write document, surf the web.
• Several application serves the same task
• OpenOffice and Word
• Standardized data format.
• Application metadata
• Window position, window size, cursor position.
• Related work Project Aura (checkpoint/resume).

15
Prototype 1 Task migration
Task Wordproc.
app1
lttask id139 typeWord processinggt ltdata
filethesis.doc/gt ltstate cursor43
windowsize32/gt .
lt/taskgt
16
Prototype 1 Task migration
lttask id139 typeWord processinggt ltdata
filethesis.doc/gt ltstate cursor43
windowsize32/gt
. lt/taskgt
17
Prototype 1 Task migration

• Lesson possible to migrate tasks smoothly once a
  standardized data format exist.
• Problem We needed to write all test application
  from scratch.
• Unrealistic in short perspective.
• Wrong approach.

18
Approach

• Investigate if and how existing applications can
  be moved
• PowerPoint, IExplorer, Word, WMP, games.
• Goal understand constraints and structure of a
  running application that is to be moved about
  (i.e. open files, resource dependencies, open
  sockets, licencing policies, security, etc)

19
Prototype 2 Wrappers

• Migrated common Microsoft applications
• Word, PowerPoint, IExplorer, etc.
• Wrapped COM objects to manipulate internal state
• Retrieve and recreate state
• State file
• Application specific data.
• Metadata
• Window position, size, etc.

20
Prototype 2 Wrappers
Wrapper
COM object
State File
21
Prototype 2 Wrappers
State File
22
Prototype 2 Wrappers
Wrapper
State File
COM object
23
Prototype 2 Wrappers

• Lesson Exporting internal application state
  allows us to migrate an application.
• Problem
• COM APIs are large and complex.
• Resulted in large and complex wrappers.
• Specialized wrapper for each application.
• To much a COM hack.
• Need Simple checkpoint/resume mechanism.

24
Prototype 3 WAIFARER

• Observation is a large body of software with a
  more convenient mechanism for checkpoint/restart.
• X-Session Management Protocol (XSMP)
• X-Consortium standard.
• Implemented by Gnome, KDE, etc.
• Migration by combining with
• State transfer.
• Small control application.

25
Prototype 3 WAIFARER
Gedit
Register Restart command gedit ./xdh1-gedit
SM
SM
Home
Work
26
Prototype 3 WAIFARER
Gedit
SM
SM
Home
Work
Global view
27
Prototype 3 WAIFARER
Gedit
SM
SM
Home
Work
28
Prototype 3 WAIFARER
Gedit
SM
SM
Home
Work
29
Prototype 3 WAIFARER
Gedit
SM
SM
Home
Work
30
Prototype 3 WAIFARER
Gedit
SM
SM
Home
Work
31
Outline

• Context and problem.
• Software mobility.
• Lessons so far.

32
Lessons (so far)

• General separation of concern principle applies
• Mobility orthogonal to application
• Possible to use existing APIs and services to
  support software mobility.
• logging, check-pointing, recovery mechanisms.
• Mobility should be a firstorder design principle
  we are washing out details concerning this
  principle now.

33
Current work

• Next step is to move other processes.
• Long-lived GRID computations.
• 24/7 servers (dynamic reconfiguration).
• WAIF Personal File System (spanning all your
  storage devices/nodes for a single user)
• Automatic replication.
• Version controlling.
• ..
• Heterogenety.
• TASK description RDF schema for TASK.

34
Last Words

• More info
• http//www.waif.cs.uit.no/
• Public domain WAIFARER http//sf.net/projects/wa
  ifarer/
• ?

35
Software mobility

• HW - Common hardware
• Laptop, PDA, cellular.
• OS - Common SW-platform
• Sprite, Mach
• App - Common application base
• TACOMA, Aglets, MOLE, .
• Task - Common data format
• Project Aura, WAIF

Task
App
OS
HW
36
WAIF Proactive Internet

• The web works autonomously on your behalf and
  notifies you.

S
p

• Goals
• High recall.
• Extreme precision.
• Context-aware.
• Real-time.

S
p
S
p
S
p
37
The Web-Service problem

• Generalized server functionality.
• Applications needs to do more work
• More interactions.
• Pull model.
• Resource locality.

C
S
WS
C
WS
WS
38
Pervasive computing problem

• Locality parameters changes.
• User movement
• Relocate deployed application components.
• Weak mobility

Device
f
Device
f
Device
f
Device
f
Device
Device
39
Solution Factorize the desktop

• Applications are resource localized.
• Migrate applications when necessary
• E.g. as a result of user movement

Messanger
Word
Nero
Virtual Desktop
40
Resource locality
to servers
Net services, storage, etc.
User interface, device access
to user