Multi-Paradigm Computing Architecture for Power Efficient Intent Based Networks

Machine learning plays an important role in next generation Intent based networks. The realization of the potential of machine learning requires big data processing. This can be achieved in cloud computing platforms that utilize Von-Neumann hardware. Von-Neumann hardware big data processing for machine learning is power intensive. Therefore, a mechanism for realizing low power big data processing and machine learning algorithm development is required. The use of neuromorphic computing hardware with low power consumption can achieve this goal. This paper proposes a multi-paradigm computing architecture that incorporates neuromorphic and Von Neumann hardware. This is done to protect existing investment in Von Neumann hardware infrastructure. The proposed architecture is intended for use in machine learning driven Intent based networks. The paper also proposes the use of a pause feature to ensure that unused processors are inactive state. Performance evaluation shows that the proposed mechanism enhances the existing approach of using only Von-Neumann hardware. The proposed mechanism reduces cloud power consumption, enhances data transmit power, number of data transmit epochs and the power usage effectiveness by up to 51.3%, 28.4%, 94% and 68.2% on average respectively.