Program function and feature
Implementing clustered system
Multiple PCs are connected via LAN, and one PC calculates the simulation. The calculation results are sent to the other PCs, which then process the images to display. In this system, the machine that calculates the simulation is called "the server" and the machines that receive the calculation results and process the images are called "slave machines". The calculation and processing of images are performed as follows.
- Simulation is calculated once for each frame
- After each calculation, the data required is sent to the slave machines
- While the slave machines process these images, the server calculates more
data for the next frame
- The server will send the data for the next one frame, if the calculation takes less time than the time it takes for the slave machines to process the image.
This optimization minimizes problems caused by data sending/receiving time. Multicast IP is utilized to send data to the slave machines, so that it minimizes the work load at the server and on the network.
- After image processing finishes, the server will perform the next calculation
and data transmission.
|http://www.forum8.co.jp/english/image/sankaku180.gifData flow by frame
In the cluster system, one server machine calculates the simulation and the slave machines are used solely to process images. Because of the distributed workload, the performance of the system is improved.
The previous versions of UC-win/Road processed all the calculations except for the traffic in a single thread, but for this multiple display channel function, the calculation and the processing had to be performed independently. As a result of optimized use of multi CPU and multi-core CPU on the standard configuration of the UC-win/Road, we can see the improved performance of 3D trees, flags, clothes, fire and smoke and the animation of MD3 characters.
| http://www.forum8.co.jp/english/image/sankaku180.gifSynchronization of flags, fire and smoke,
vehicles and snow
The data, which server sends to the slave machines, are the information
and attributes of moving elements. Each slave machine processes images
after receiving data to update the 3D space. When visualizing 3D trees,
flags, rain and snow, the processing becomes more complicated, thus, the
data becomes larger, and therefore, high speed data transmission over a
network becomes nearly impossible. Hence, to process of these objects,
server sends minimum parameters for the calculation and the slave machines
processes the calculation.
As the functions for the cluster system was added with the standard version of UC-win/Road as the basis, the system has all the features of the standard UC-win/Road. To minimize operations on the slave machines, all the operations except for the initial settings can be done on the server machine. At the time of simulation, operation is done by server machine and visualization is done by slave machines.
The cluster system is operable only when using the following functions with following settings.
|Initial setting of slaves
||slave machine name, network address of server
||Area displayed by slave machines
|Disconnect slave machines
|Restart, shut down and data import of slave machines
|Switching to server/ standard mode or automatic start of server mode
|Simulation start/ end
We measured the level of performance to check the validity of this system.
We found that its performance improved, in comparison with the standard
UC-win/Road system, and that the number of clustered PC does not affect
the performance. In addition, by comparing 4-channel clustered configuration
with the standard configuration, we saw the improvement of 140%. By comparing
15 slave machines and standard 1 channel, we saw the improvement of 20
cluster configuration, how to process separately