Next: Conclusions Up: Progressive Iso-surfaces on the Previous: The Java User Interface

Results

 
Table 1:   Visualization of levels 0-4 (LAN)

 
Table 2:   Refinement from level 5 to highest level (LAN)

 
Table 3:   Visualization of levels 0-4 (Berlin-Erlangen)

 
Table 4:   Refinement from level 5 to highest level (Berlin-Erlangen)

The performance of the transfer and visualization of the surface levels for an iso-surface through a LAN and WAN net is the most important issue. We made tests on two different types of client machines and two classes of network connections.

The client machines were a SGI O2 with a R10000 processor and 128 MB RAM and a PC with a 166 MHz Pentium processor and 64MB RAM. We used Netscape Communicator 4.04 with the JDK1.1 preview 2 patch. The server was a SGI O2 with a R10000 processor and 128MB RAM.

For the measurements we used three different data sets. The first data set represents a spherical harmonic function discretized on an uniform grid with 16 bit precision. The second data set is a MRI head with samples and 16 bit precision. The third data is a numerical simulation of the fluid motion around a wing. The data is discretized on a structured grid of dimension . The size of the data set is 4.2 MBytes. The data sets were pre-processed and stored in compressed format as described in [4]. The sizes of the compressed data sets are 900 KBytes for the spherical harmonic data set, 1.1 MBytes for the MRI head data set and 3.2 MBytes for the wing data set.

A local area network and a Berlin-Erlangen backbone Internet connection (100 MBit) were tested. The distance between Berlin and Erlangen is about 450 Km. The benefits of our approach are obvious. The times given include the network transmission and the construction of the VRML scene graph. On the LAN as well as on the remote Internet connection, the levels 0 to 4 of the iso-surfaces were visualized in less than one second (Table 1 and 3). The user is able to get a quick impression of the data and find interesting iso-values.

The performance for the refinement of the iso-surface up to the highest level of detail is obviously much lower (Table 2 and 4), but still high enough to maintain an acceptable rate of interactivity.

The bottleneck of the whole application is the network transmission. This can be seen in the following example. The refinement of the MRI head data set from level 5 to 6 takes 15.1 sec. which can be split in 14,7 sec for the network transmission and 0,4 for the geometry processing. The number of triangles transmitted in this case is 45,884.