A Low-Cost Energy-Efficient Raspberry Pi Cluster for Data Mining Algorithms

João C Saffran (1) , Gabriel C Garcia (1) , Matheus Souza (1) , Pedro Henrique Penna (2) , Marcio Castro (3) , Luís F. Góes (1) , Henrique C Freitas (1)

1 Pontifical Catholic University of Minas Gerais [Belo Horizonte]
2 PUC Minas - Pontifícia Universidade Católica de Minas Gerais
3 UFSC - Universidade Federal de Santa Catarina = Federal University of Santa Catarina [Florianópolis]

Data mining algorithms are essential tools to extract information from the increasing number of large datasets, also called BigData. However, these algorithms demand huge amounts of computing power to achieve reliable results. Although conventional High Performance Computing (HPC) platforms can deliver such performance, they are commonly expensive and power-hungry. This paper presents a study of an unconventional low-cost energy-efficient HPC cluster composed of Raspberry Pi nodes. The performance, power and energy efficiency obtained from this unconventional platform is compared with a well-known coprocessor used in HPC (Intel Xeon Phi) for two data mining algorithms: Apriori and K-Means. The experimental results showed that the Raspberry Pi cluster can consume up to 88.35% and 85.17% less power than Intel Xeon Phi when running Apriori and K-Means, respectively, and up to 45.51% less energy when running Apriori.

Mots clés: Raspberry Pi cluster; Intel Xeon Phi; Apriori; K-Means

Domaines: Performance et fiabilité [cs.PF]; Calcul parallèle, distribué et partagé [cs.DC]

João C Saffran, Gabriel C Garcia, Matheus Souza, Pedro Henrique Penna, Marcio Castro, et al.. A Low-Cost Energy-Efficient Raspberry Pi Cluster for Data Mining Algorithms. European Conference on Parallel and Distributed Computing Workshops (Euro-Par Workshops), Aug 2016, Grenoble, Rhône-Alpes, France. ⟨hal-01506931⟩