1.   Strategies and Techniques for Fighting the Bandwidth 'Bottleneck'

1.1.  First of all, let's look at some interesting figures extracted from one of the NAS Technical Reports - 'Scientific Visualization in High-Speed Network Environment', Vaziri A. NASA Technical Report.
 

Figure 1 The NAS Processing System Network (NPSN)  Figure 2 The physical connectivity of the NPSN 
Figure 3 CFD simulation and visualization cycles Figure 4 Example of post-processing CFD graphic 
Figure 5 Single frame of Ultra image Figure 6 NPSN network logical layout target configuration 1991
Figure 7 the UltraNet test configuration  Figure 8 Direct volume rendering of turbulent boundary layer
Table 1 Specifications for Workstation 2  Table 2 CFD problems and applications on NPSN
Table 3 Characteristic size of CFD datasets Table 4 Characteristic size of visualization datasets
Table 5 UltraNet transfer rates in kbps Table 6 Projected technical evolution of simulation/visualization environments

 

1.2. "Adapting to Bandwidth Variations in Wide-Area Data Combination", Ranganathan M., Acharya A. and Saltz J., National  Institute of Standards and Technology, Gaithersburg, MD. Technical Report.

This paper describes an approach for transferring volume image data by dapting to large variations in available network bandwidth based the fact that the communication cost dominates the total system response time in the Internetwork environment. The implementation is a system of many 'operators' (servers) communicating with a central server to, which dynamically retrieves images from various geographically sites and combine them together before delivering to the intended users. Three algorithms are developed and implemented to monitor local and global network bandwidth status in order to make intelligent decisions (adapting a combination plan). The network bandwidth status is constantly monitor and the system is receiving these information as inputs for calculating the critical path for an image to be quickly delivered.

(1) 'One-Shot' Algorithm: is initiated at the begining of a session to decide which 'operator' takes the job.

(2) 'On-line' local algorithm runs at real time to dynamically reposition the operators. the 'one-shot' algorithm is used to calculate a 'good' placement based on the
      local information between two hosts.

(3) 'On-Line' global algorithm is run periodically and currently with the actual computation. it also uses the 'one-shot' algorithm to calculate the initial placement to
      calculate the new placements. If a change-over should take place, the message will be braodcast to all the servers.