Parallel Computation Techniques for Ontology Reasoning

Abstract

Abstract. As current reasoning techniques are not designed for massive parallelisation, usage of parallel computation techniques in reasoning establishes a major research problem. I will propose two possibilities of applying parallel computation techniques to ontology reasoning: parallel processing of independent ontological modules, and tailoring the reasoning algorithms to parallel architectures. MotivationScalability is an issue that is subject in many semantic web research discussions. More and more researchers share the awareness that reasoning in its current form will not be able to bear the load of data it is supposed to handle in the near future. A polarising article was published by Fensel and van Harmelen [1], where the authors talk about "10.000 triples just to describe each human, which gives us 100 trillion." Even though this guess may be intensionally provocative, it has in fact been proven several times in the past that even high estimations of growth were beaten in reality often long before they were predicted to eventuate.An oberervation is, that available state-of-the-art reasoners do not exploit the benefits of parallel computation techniques, as these are not straightforwardly applied for reasoning calculi. Multithreading or other ways of distributed computation cannot easily be taken care of by the operating system. This is a major problem, since computational power at the level of integrated circuits is about to explore its physical limits by being "down to atoms" concerning conductor and transistor size. However, parallel computer architectures emerge, such as grid, peer-to-peer, or multi-core machines even in home-computing environments. This allows for the overall computational power to grow further, provided that software architecture and algorithms respect this computational paradigm shift. Thus parallel architectures do not inherently speed up all kind of computation, as the workload needs to be split into chunks of independent computations. Current reasoning algorithms do not naturally decompose into independent computational chunks.In fact, it remains an open question whether algorithms currently used in reasoning adapt well to the paradigm shift in computer architecture. In particular it is unclear whether well established tableau algorithms, as widely used in stateof-the-art reasoners, can be parallelised.

Ontology Design for Solving Computationally-Intensive Problems on Heterogeneous Architectures

“…An integrated solution of ontological reasoning and parallel computing to address issues of scalability is presented in [13]. Parallel computer architectures, such as home computing environments, multi-core machines, grid and peer-to-peer, lead to great demand for efficient handling of computational resources through making use of software architecture, algorithms and now a new paradigm-making use of ontological reasoning.…”

“…Parallel computer architectures, such as home computing environments, multi-core machines, grid and peer-to-peer, lead to great demand for efficient handling of computational resources through making use of software architecture, algorithms and now a new paradigm-making use of ontological reasoning. In [13], the authors propose using ontology modularization and queries. This helps to solve the problem of reasoning in individual modules, ultimately to achieve maximum efficiency in the use of architectures and parallel computing algorithms.…”

“…For example, a framework as an integrated solution of parallel computing and ontological reasoning is presented in [13]. It can be used on scalability issues in parallel computing by using ontological reasoning.…”

Parallel Materialization of the Finite RDFS Closure for Hundreds of Millions of Triples

“…Less related works include reasoning over distributed hash tables [9,10]; potentially parallel evolutionary query answering [11]; composition of approximate, anytime reasoning algorithms executed in parallel mentioned at the end of [12]; proposals of parallel computation techniques for ontology reasoning [13]; and approaches for exploiting vertical parallelism in tableaux-based description logic reasoning [14] .…”

“…Similarly, as the size of the data set doubles, the time to inference doubles. 13 On 128 processes, LUBM10k/1024 takes 1.10 seconds, and LUBM10k/4 takes 291.46 seconds. Figure 2 shows that the overall time of the computation is generally linear, but as the smallest data set is run on a larger number of processes, the speedup decreases.…”