Frank Paterra

Distributed simulation: No special tools required

A toolkit of C language functions that can be linked with SIMSCRIPT programs to provide the data communication primitives necessary for distributed simulation is presented. This toolkit can be used to develop distributed simulation applications without having to invest in new environments or training. The toolkit is composed of 650 lines of C code and requires about 50 lines of additional SIMSCRIPT code per server/receiver pair to be added to the simulation model. The user need only be concerned with three C function calls and two SIMSCRIPT routines; thus the complexity of the simulation program is no significantly affected. The use of these tools is most likely to be beneficial in models that can be decomposed into components in which information flow among processors is not cyclic.<<ETX>>

Remote visualization and parallelism using existing networks

A method for using existing networks and available parallel channels to provide the needed throughput for remote visualization is proposed. This method keeps the functionalities of the traditional protocol stack but introduces parallelism at all points where bottlenecks can develop. The approach allows the use of existing protocol standards and off-the-shelf hardware to permit new applications that require higher bandwidths. The design tradeoffs in this approach are discussed to develop working architectures needed to solve the animation problems of 5, 10, and 33 frames/s, or 10 Mb every 200 ms, 100 ms, and 30 ms, respectively.<<ETX>>

Concurrent use of parallel communication to enable remote visualization

Visualization of scientific data is the subject of much recent research. The computational power needed to produce animated sequences in near real time is great and, while not locally available to many research facilities, can be provided remotely. Significant bandwidth is required to deliver animated images in real time, and is normally not available without dedicated network lines. This paper investigates the requirements needed to deliver remotely computed data via multiple, existing networks, used in parallel. The authors present a first order estimation of the protocol processing required to reliably transmit the images, and simulation results showing expected throughput available under several communication and network management policies.<<ETX>>

Utilization of special architectures in support of parallel discrete event simulation

The authors describe a project which is intended as a prototype study to demonstrate the effectiveness, as measured by execution speedup, of running parallel simulations on newly available high-performance but relatively low-cost multiple-processor architectures. The authors focus on performance problems related to the use of discrete event simulation as a tool for development of system designs, and they discuss important constraints in its use in this area. They briefly survey the high-performance architectures which can be used to speed up simulations, as well as the relationship of parallel simulation to general distributed systems issues, and they describe the model being implemented to evaluate the effectiveness of the architecture and processor synchronization technique that have been chosen.<<ETX>>