These methodologies address power-related challenges, such as the use of current and future devices or the scheduling of virtual machines across data centers, among others.
Regarding programming models, the project has focused on shifting from a shared-memory model to one centered on actors and message exchange. This allows programmers to achieve greater energy efficiency while simultaneously increasing their awareness of energy-related issues. Two illustrative approaches are:
• Individualized programming solutions for heterogeneous GPU/CPU architectures. Managing the GPU directly through optimized code can reduce energy consumption by approximately 80%. However, this requires specialization that reduces programmability. The ParaDIME project has developed techniques based on domain-specific languages ​​that generate code for both the CPU and the GPU, enabling energy savings of up to 40% and, crucially, making these architectures more accessible to a wider range of programmers.
• Tools that promote greater awareness of power and costs. These tools assess the power requirements of a single process running in a virtualized environment. They can also be used for user-centric cost models, to implement task scheduling that considers energy consumption, and as an indicator of the amount of heterogeneous resources consumed by an application.
Regarding execution time, the ParaDIME project has developed a large, decentralized infrastructure consisting of small data centers that provide heating and hot water. This system would lead to increased efficiency, as demonstrated by the project's industrial partner, the German company Cloud&Heat. ParaDIME researchers have developed:
• A multi-data center scheduler: This schedules tasks across different data centers, achieving a balance between data center loads and heating/cooling needs. This has resulted in up to a 50% reduction in CO2 emissions and energy consumption.
• A cross-data center scheduler: Technologies have been developed to reduce both the time required to reactivate virtual machines and the costs associated with their migration. The QEMU community (an open-source virtual machine emulator and emulator) is analyzing parts of this work. Institutions currently using QEMU to virtualize their workloads will benefit from migration code for virtual machines optimized by the ParaDIME project. In addition, the ParaDIME project has contributed a feature that tracks changes to block devices, which has already been incorporated into the latest Linux kernel.
At the hardware level, ParaDIME project researchers have proposed and simulated various methodologies to improve compute node efficiency, including:
• Scheduling tasks across heterogeneous cores (e.g., big.LITTLE processors, or systems that combine FPGA, GPU, and CPU cores). An average of 40% energy consumption can be saved by combining FPGA, GPU, and CPU cores compared to a multicore processor. ParaDIME scheduling also reduces power consumption by an average of 20% and energy consumption by 30% across different types of heterogeneous platforms. The ParaDIME project has also analyzed power estimation tools for a variety of cores.
A significant reduction in supply voltage is achieved. Energy savings are obtained by combining this reduction with low-cost error detection and correction techniques. Furthermore, ParaDIME researchers have studied this methodology in circuits built with future devices. The ParaDIME methodology saves up to 60% of the energy consumed by the L1 data cache.

More information