New hardware mixing, multicore CPU, GPU and FPGA on chip, on board and in server clusters are providing increasing power while also making it possible to reduce drastically the energy consumption. However these new heterogeneous hardware architectures are notoriously hard to program. The Horizon2020 project TANGO aims at facilitating the optimal use of hardware with heterogeneous computing and networking capabilities. CETIC brings its expertise from ASCETIC (Energy Efficient Clouds) and POLCA (exploitation of program data flow to increase parallelism).
TANGO kickoff gathered the 6 consortium partners of which 2 are from Wallonia (CETIC and Deltatec) in Madrid, Spain, home of the Atos research unit that coordinate the project. A small executable prototype is already planned for the end of 2016. Prior to jumping in the implementation, the initial task consists of requirement elicitation on the business, as well as on the technical, side. A review of the state of the art on various relevant subjects of interest to TANGO is also on-going. In this context, CETIC is currently reviewing research on software+hardware modelling, and on optimal scheduling techniques.
Furthermore, during the kickoff, Deltatec in collaboration with CETIC presented one of the two use cases that will be performed during the project. This case will exploit the power of FPGA as well as clusters of CPU+GPU in order to make image processing and machine learning algorithms perform faster with reduce energy consumption rate. These algorithms will, for instance, be of interest in domains such as TV broadcasting or eHealth.