An enhanced simulated annealing approach for cylindrical, rectangular mesh, and semi-diagonal torus network topologies

A multiprocessing system has processor-memory modules in a network which is always referred to as net. In many cases, the modules are placed in a regular arrangement such as rectangular grid, bus, star and hypercube. In this research, we proposed one conceptual model and two network topologies for routing the elements of the network. In the first model, a static single-row network was transformed into a dynamic three-dimensional cylindrical model. This new routing model has its axis perpendicular to single-row planes, which gives the advantage of allowing unlimited connections between the pairs of elements based on the program requirements. The single-row routings in each network were produced optimally using the earlier model called Enhanced Simulated Annealing for Single-row Routing (ESSR). In the second part of this research, mesh network topology which consists of an array of square cells was proposed as our routing platform to achieve a complete automatic routing. The problem was further split into two cases; first, a fully gridded network to minimize the number of layers and second, the obstacle avoidance network model. Dijkstra?s shortest path algorithm was used to provide the shortest path for each net. The arrangement was further refined using a simulated annealing method. From this technique, the minimum number of layers was produced to complete the routing with lower energy level and to provide the best path if it exists, with the presence of obstacles. The last part of this research is an extension of our previous work, where a more scalable and regular network called semi-diagonal torus (SD-Torus) network was used as a routing platform instead of the mesh network. The performance of SD-Torus network was much better compared to torus and mesh networks in terms of energy level and the number of routed nets. The network topology performed complete routing up to 81 nodes with 80 nets in 9?9 network size. This technique maximizes the number of nets through the minimum energy. The simulations for each network are developed using Microsoft Visual C++ 2010 programming language