Improvement and extension of a middleware system for Big Data analytics

Takuya Araki (NEC Corporation)

We are conducting the research and development of a middleware system called Feliss for Big Data analytics on vector architectures. In this talk, Two recent updates of the development will be explained. One is the performance improvement of machine learning algorithms on sparse datasets. It reduces the amount of communication by utilizing the sparsity of the data. The other is the extension of the middleware that enables communication with Spark on a x86 cluster. With this extension, users can utilize rich I/O functionalities of Spark together with the high performance machine learning implementation of Feliss. The future extension of the system will also be mentioned.