MC64-Cluster: Many-Core CPU Cluster Architecture and Performance Analysis in B-Tree Searches

Abstract

The MC64-Cluster computer platform was designed, based on many-core CPU microprocessors: Tile64. MC64-Cluster architecture was outlined in terms of both hardware and software, including commands available to manage jobs and provided application programming interfaces to communicate and synchronize tiles, making this system easy to use. Massively, concurrent searches of keys in B-trees, which are used in many applications, including bioinformatics, were used. Remarkable performance improvements were obtained when the cluster resources were combined with those available in host machine (hybrid or heterogeneous environments). These results were even more outstanding when analyzed in terms of performance-per-watt, highlighting their green-computing advantages. Together with the cluster architecture, they represent the main contributions of this work. To our knowledge, this is the first cluster implementation of this kind being developed.

BLVector: Fast BLAST-Like Algorithm for Manycore CPU With Vectorization

“…BLVector is designed specifically to use the Single-Instruction Multiple Data (SIMD) vector instructions of the Advanced Vector eXtensions (AVX)-512 instruction set, as well as its multithread capabilities. We have previously demonstrated the usefulness of optimizing bioinformatics algorithms to significantly enhance performance ( Gálvez et al, 2010 , 2016 ; Díaz et al, 2011 , 2014 ; Esteban et al, 2013 , 2018 ).…”