The paper presents an analysis of the suitability of the architecture of dynamic SMP clusters with communication on the fly for massively parallel fine grain numerical computations. It is assumed that the proposed architecture is implemented using the highly modular " system on chip" and "network on chip" technology. This technology is considered to provide soon a very large number of co-operating processors embedded in a single parallel system, thus enabling massively parallel computations. The proposed architecture of dynamic clusters with communication on the fly meets requirements of large scale fine grain computations and can be successfully applied in this technology. Experimental simulation results are presented concerning efficiency of fine grain parallel implementation of a typical numerical problem which is matrix multiplication based on recursive data decomposition. Selection of optimal parallel computation grain is discussed. Estimations of the efficiency of the proposed methods for fine grain computations for large problem size are presented.
6 November 2018 in Concurrency and Computation: Practice and Experience
The paper concerns program scheduling methods oriented towards "System on Chip" (SoC) -based modular parallel architectures with efficient features of inter-processor communication. In the assumed architecture, a global communication network connects many SoC modules in which SMP clusters are dynamically created at program run-time to provide transfers of shared data on the fly for many processors at a time. Programs are represented as extended macro data flow graphs, which adequately describe communication between processor data caches and shared memory modules. The proposed scheduling algorithm is composed of two phases. The first phase distributes program graph nodes among SoC modules, assuming full inter-processor connection networks. It is implemented as a genetic algorithm with internally embedded ETF heuristics. The second phase of the algorithm schedules computation and communication inside SoC modules, so as to optimally use dynamic processor switching between clusters and data read on the fly mechanisms. Scheduling results of sample program graphs evaluated using simulation methods illustrate the efficiency of the proposed algorithm.
Expressing attack-defence trees in a multiagent setting allows for studying a new aspect of security scenarios, namely, how the number of agents and their task assignment impact the performance, e.g. , attack time, of strategies executed by opposing coalitions. Optimal scheduling of agents' actions, a nontrivial problem, is thus vital. We discuss associated caveats and propose an algorithm that synthesizes such an assignment, targeting minimal attack time and using the minimal number of agents for a given attack-defence tree. We also investigate an alternative approach for the same problem using rewriting logic, starting with a simple and elegant declarative model, whose correctness (in terms of schedule's optimality) is self-evident. We then refine this specification, inspired by the design of our specialized algorithm, to obtain an efficient system that can be used as a playground to explore various aspects of attack-defence trees. We compare the two approaches on different benchmarks.