Faculty: Computer Science and Technology
Speciality: System Programming
Parallel computing attract manufacturers of their accuracy and speed compared with conventional systems. A small number of possible architectures and the high cost of preventing to use them everywhere. Therefore, usually as a parallel system using a multicore computer or a cluster of several of the same type of desktop computers. Typically manufacturing systems require solutions of similar or identical tasks. Programmers have to create special software products that run on a particular architecture. It was suggested to use as a parallel system based on FPGA device because of the possibility of reprogramming.
The urgency of the work is confirmed by the growing interest in cheapening of parallel systems and the creation of affordable solutions for everyone in the field of parallel computing and simulation. There is growing interest in the possibilities of hardware implementation of various algorithms, creation of specialized systems designed for rapid solution to a problem with a certain accuracy. At the moment there is no software and hardware that can simplify the implementation of the software and hardware for a parallel system. This leads to the fact that manufacturers are using other people's computing power or reject them.
The goal of master's work is research and develop methods of automated design and creation of parallel machines based on a dynamically tunable circuits FPGA. The main problems are:
Alleged scientific novelty consists in the development of algorithms for mapping virtual network models of dynamic objects and algorithms for the creation of specialized parallel machines based on chips with reconfigurable logic such as FPGA.
As a practical results will be to get some device with a reconfigurable architecture based on FPGA for solving systems of differential equations that describe the network of dynamic objects and obtain comparative characteristics of the results of parallel supercomputers.
There are researched the network of dynamic objects and ways to mapping virtual parallel models on the existing architecture. The main shortcomings of researched methods is focuses it on their specific architecture and the impossibility to using them for a dynamically changing architecture. Figure 1 shows the classic way to conditionally mapping the model on the existing architecture. As can be seen from the figure, leads to some problems at the level of architecture because of its limitations or inability to the realities of a particular task.
The capability of dynamically rebuilding the internal architecture of chips Spartan-3E, conducted researches on the opportunities of the dynamic reprogramming unit chips are used as microprocessors, as well as the dynamic extension of the device, built on several types of chips Spartan-3E
Thus, as the desired results and the proposed targets is the achievement of fashion design, shown in Figure 2, in which the development of hardware and software are complementary and mutually exclusive processes, carried out in parallel at the level of development of modeling (critical) environments.
Since the problem of simulation of network dynamic objects are widely used and well known, it is planned to create an extensible and dynamically changing environment modeling, changing either automatically or with minimal human influence, under the necessary tasks. The optimum is to build a system with the ability to solve several tasks at one device by using chips that do not participate in the computation of other tasks (located in a prime).