Radu Prodan presented the paper “Proactive SLA-aware Application Placement in the Computing Continuum” at the 37th IEEE International Parallel & Distributed Processing Symposium (IPDPS 2023) <https://www.ipdps.org/> in St. Petersburg, Florida, USA.

Authors: Zahra Najafabadi Samani, Narges Mehran, Dragi Kimovski, and Radu Prodan, Alpen-Adria-Universität Klagenfurt

Abstract: The accelerating growth of modern distributed applications with low delivery deadlines leads to a paradigm shift towards the multi-tier computing continuum. However, the geographical dispersion, heterogeneity, and availability of the continuum resources may result in failures and quality of service degradation, significantly negating its advantages and lowering users’ satisfaction. We propose in this paper a proactive application placement PROS method relying on distributed coordination to prevent the quality of service violations through service-level agreements on the computing continuum. PROS employs a sigmoid function with adaptive weights for the different parameters to predict the service level agreement assurance of devices based on their past credentials and current capabilities. We evaluate PROS using two application workloads with different traffic stress levels up to 90 million services on a real testbed with 600 heterogeneous instances deployed over eight geographical locations. The results show that PROS increases the success rate by 7-33%, reduces the response time by 16-38%, and increases the deadline satisfaction rate by 19-42% compared to the two related work methods. A comprehensive simulation study with 1000 devices and a workload of up to 670 million services confirms the scalability of the results.

As part of our collaboration with the Department of Computer Architecture and Technology at the University of Granada, Spain two research papers were accepted for publishing at the 17th International Work-Conference on Artificial Neural Networks.

Paper title:  An Efficient Parallel Multi-population Wrapper for Solving Feature Selection Problems in High-dimensional Space

Authors: Juan Carlos Gómez López, Daniel Castillo Secilla, Dragi Kimoviski and Jesús González Peñalver

Abstract: One of the most widely accepted approaches to address feature selection problems are wrappers based on evolutionary algorithms. Over the years, these approaches have evolved from single-population models to multi-population models to achieve better-quality solutions. Moreover, they are highly parallelizable as each subpopulation evolves independently. This paper proposes two parallel strategies for a multi-population wrapper to take advantage of a multicore CPU. The first one, based on the Fork/Join model, focuses on parallelizing only the evaluation method since it is the main bottleneck of the procedure. Although this strategy speeds up the execution of the wrapper, it is far from optimal. In this context, the second strategy implements an Island-based model, where each subpopulation evolves independently in each CPU core, exchanging information via asynchronous migrations. The results show that the wrapper achieves a speedup of almost 35 with the Island-based model when individuals are distributed into 24 subpopulations.

Paper title: Energy-aware KNN for EEG Classification: A Case Study in Heterogeneous Platforms

Authors: Juan José Escobar Pérez, Francisco Rodríguez, Rukiye Savran Kızıltepe, Beatriz Prieto, Dragi Kimovski, Andrés Ortiz, Alberto Prieto and Miguel Damas

Abstract: The growing energy consumption caused by IT is forcing application developers to consider energy efficiency as one of the fundamental design parameters. This parameter acquires great relevance in HPC systems when running artificial neural networks and Machine Learning applications. Thus, this article shows an example of how to estimate and consider energy consumption in a real case of EEG classification. An efficient and distributed implementation of the KNN algorithm that uses mRMR as a feature selection technique to reduce the dimensionality of the dataset is proposed. The performance of three different workload distributions is analyzed to identify which one is more suitable according to the experimental conditions. The proposed approach outperforms the classification results obtained by previous works. It achieves an accuracy rate of 88.8% and a speedup of 74.53 when running on a multi-node heterogeneous cluster, consuming only 13.38% of the energy of the sequential version.