Distributed Post-processing and Rendering for Large-Scale Scientific Simulations - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 372

•

Network Latency: The delay between sending an interaction command, e.g. a
viewpoint position, and receiving frame updates can lead to unacceptable convul-
sive movement of objects.

• Network Bandwidth: The bandwidth requirements for sending large polygonal

data or series of high resolution images between local and remote resources can
be very high.

• Error Recovery: Errors while transferring data can lead to unacceptable visual

artifacts or poor application behavior.

• Scalability: Most of the workload is shifted to remote resources. The number

of concurrently supported users is limited by network bandwidth and compute
capacities.

• Network Robustness: The system should be failure-resistant to network problems

such as packet loss.

• Load Balancing: Optimal balancing of the visualization pipeline between available

local and remote resources is not trivial.

31.3.2.1 Classification of Remote Rendering Techniques

Remote rendering techniques can be classified based on their distribution of the
visualization pipeline as depicted in Fig. 31.9 . This leads to different data types
transferred between the local and the remote resource.
A common and simple remote rendering approach is image streaming. The image
is rendered on a remote machine for the requested viewer position. Finally, the client

Abstract
Visual
Objects

Raw
Data

Derived
Data

Image
Data

Visualization
Pipeline

Filtering Mapping Rendering Display

Image-based Remote Local

Remote Local

Model -

Based

Remote Local

Remote Local

Hybrid Remote Local

Fig. 31.9 Classification of distributed rendering techniques based on their pipeline distribution.
Depending on the chosen technique some pipeline stages are executed on local resource and others
on remote resources