During the last years, deep learning (DL) models have been used in several applications with large datasets and complex models. These applications require methods to train models faster, such as distributed deep learning (DDL). This paper proposes an empirical approach aiming to measure the speedup of DDL achieved by using different parallelism strategies on the nodes. Local parallelism is considered quite important in the design of a time-performing multi-node architecture because DDL depends on the time required by all the nodes. The impact of computational resources (CPU and GPU) is also discussed since the GPU is known to speed up computations. Experimental results show that the local parallelism impacts the global speedup of the DDL depending on the neural model complexity and the size of the dataset. Moreover, our approach achieves a better speedup than Horovod.

Driven by Big Data constraints, several new database paradigms have emerged, one of them is the Wide Column Store (WCS) paradigm. Current WCS solutions are relatively mature, but they have many drawbacks. This project aims to provide a WCS alternative by combining the WCS paradigm with the Relational DataBase Management System (RDBMS) architecture, which is called: the Relational Wide Column Store paradigm. Early results show a 31 % performance increase with read queries and relatively solid writing performances. This early-stage paper is encouraging and demonstrates that there is still room for improvements with RDBMS systems.