eViz

1
e-Viz

• Towards an Adaptive Framework for Visualization
  on the Grid

2
e-Vizzers

• e-Viz is a three year joint research project
  funded by UK EPSRC
• Four partner universities
• University of Leeds
• (Ken Brodlie, Jason Wood)
• University of Manchester
• (John Brooke, Mark Riding)
• University of Wales, Bangor (Nigel John, Chris
  Hughes)
• University of Wales, Swansea
• (Min Chen, David Chisnall, Mark Jones, Nicolas
  Roard)

UW Bangor
U Manchester
U Leeds
UW Swansea
3
What problem are we trying to solve?
Which are applicable to the data?
Which is the most suitable?
Whats fastest? Availability?
Data
Choose Visualization Algorithm
Choose Software to Implement Algorithm
Choose Hardware to run on
Visualization Specialist
e-Scientist
Domain Scientist
4
Why are we trying to solve this?

• Because not all potential Grid visualization
  users are experts in visualization and Grid
  technologies.

5
Why are we trying to solve this?
Grid Visualization Developers
e-Scientists
Grid Experience
Application Scientists
Visualization Specialists
Visualization Experience
6
Putting Visualization on the Grid

• Can the Grid help? its supposed to be
• heterogeneous in architecture
• seamless and transparent in use
• fault tolerant in operation
• capable of adapting to the changing environment
  in order to provide the best service.

7
What can the Grid offer?

• Heterogeneity architectures

8
What can the Grid offer?

• Heterogeneity operating systems

9
What can the Grid offer?

• Heterogeneity visualization software

10
What can the Grid offer?

• Heterogeneity Implications
• There are a wide variety of architectures,
  operating systems and software applications used
  in visualization.
• Each has its own particular rewards and
  benefits, but we cant expect every Grid user to
  be knowledgeable and proficient with each, or
  even to know which is the most appropriate for a
  given task.
• Want to be able to offer the power afforded by
  each application to all potential users of
  visualization.

11
What can the Grid offer?

• Seamlessness and transparency Implications
• End users should only have to learn one
  interface, but still be able to benefit from the
  features offered by a wide range of applications.
• But each visualization application has its own
  user interface
• and every operating system looks and behaves
  differently

12
What can the Grid offer?

• Fault tolerance Implications
• A distributed system is one in which the failure
  of a computer you didnt even know existed can
  render your own computer unusable Leslie
  Lamport
• A Grid system is one in which the failure of a
  computer you didnt even know existed goes by
  unnoticed
• Users should be able to rely on the Grid

13
What can the Grid offer?

• Fault Tolerance - User Scenario

keyboard, please
14
What can the Grid offer?

• Adaptation Implications
• The world is dynamic and ever changing so is the
  Grid.
• Network loads and status
• Queues on HPC machines
• Runtime limits on HPC machines
• CPU loading, both remotely and locally
• A Grid system should adapt to cope with changes

15
How can we solve this problem?
16
What is Visualization?
17
What is Visualization?
18
What is Visualization?
19
What is Visualization?
20
Common Abstract Interface
21
Common Abstract Interface
22
Common Abstract Interface
23
Common Abstract Interface
24
Common Abstract Interface

• Feasibility
• can we create equivalent visualizations using
  different software packages?

25
Equivalent Visualizations?
26
Equivalent Visualizations?
27
Abstraction

• Need to describe the pipeline itself need an
  abstract visualization description language
• gViz project gives us skML

28
Abstraction

• Format Conversion
• Most visualization software applications can
  already read a wide range of data formats

29
Abstraction

• Computational Steering

30
Abstraction

• Computational Steering
• RealityGrid or gViz APIs can be used to control a
  running pipeline
• Visualization applications must be instrumented
  to expose their steerable pipeline parameters
• Currently instrumented VTK and RTRT (real time
  ray tracer)
• GUI created, which is really a specialised
  Computational Steering client
• Reads in a pipeline description and dynamically
  configures itself to show appropriate widgets

31
Abstraction

• Frame Transport

32
Abstraction

• Frame Transport
• Images can be rendered locally or remotely in
  either case, they need to be displayed on the
  users own machine, as well as any collaborators
• Have created a library to compress remote images
  and transport them back to the client for
  display. Visualization applications can be
  modified to use the library
• Range of image compression codecs
• Local rendering supported via a rendering overlay
• Linux and Windows clients

33
What do we still need?

• Decision making
• What visualizations are applicable for a given
  input data type
• Which of the available hardware and software is
  most suitable to implement such a pipeline
• User may not know the answer to these questions
  provide assistance

34
Architecture
WSRF
Globus
gViz
e-Viz
35
Architecture (Current)
WSRF
Globus
gViz
e-Viz
36
Demonstration 1

• gViz Pollution Demonstrator - shows
• Active simulations as a data source
• Computational steering of an active visualization
• Common interface to servers

37
Demonstration 1
38
Demonstration 2

• Volume Render Demonstrator shows
• Heterogeneous access to multiple servers
• (Almost) seamless switching between servers
• Fault tolerance through redundancy

39
Demonstration 2
40
The Knowledge Gap

• The Broker has a knowledgebase which it will use
  to make informed decisions on pipeline choices.
• Its empty!

41
How to Distribute?
42
How to Distribute?
43
How to Distribute?
44
How to Distribute?
45
How to Distribute?
46
When to Distribute?

• Depends on
• Where the dataset resides
• How big the dataset is
• Network bandwidth
• Available machines
• Chosen algorithm
• Queue status
• CPU capabilities
• Graphical capabilities
• ..

47
Other problems

• Single point of failure in the broker

48
Conclusions

• e-Viz has created a framework for visualization
  on the Grid that is
• Heterogeneous in architecture
• Seamless and transparent in use
• Fault tolerant
• Adaptive (potentially)
• But much more to be done! (project runs another
  16 months)

49
For more information

• See a live demo
• Tuesday 22nd, 1400-1630, White Rose Grid stand
• Project Web Site
• http//www.eviz.org
• email
• mark.riding_at_manchester.ac.uk