M²DC Project Page

To achieve this objective, M2DC is built on three main pillars:

• [Pillar 1] A flexible server architecture that can be easily customised, maintained and updated so as to enable adaptation of the data centre. This architecture will be optimized with respect to the integration of computing resources with constrained thermal power dissipation such as embedded CPUs and GPUs, FPGAs, manycore processors, while also being able to integrate more powerful resources if needed. Built on established standards such as COM Express, the server architecture will enable the integration of third-party boards and chips, leading to what can be called an open server architecture. Particular attention will be put on maintainability and reliability by design, relying for example on computing resource hot plug management and redundancy of power supplies.
• [Pillar 2] Advanced management strategies [Pillar 2a] and system efficiency enhancements (SEE) [Pillar 2b]. In order to improve the behaviour of the system during runtime and to meet requirements from the various applications, the server architecture will include built-in enhancements (e.g., for computing acceleration, enhancements of the global efficiency thanks to data management, dependability and security, behaviour monitoring, etc.) on system level. High power efficiency will be provided by an intelligent power management enabling to constantly optimize the power consumption of the system thanks to proactive and reactive strategies.
• [Pillar 3] Well-defined interfaces to surrounding software ecosystem will allow for an easy integration into existing data centre management solutions through the use of the latest middleware software for resource management, provisioning, etc. Building on the Linux operating system (e.g., Linaro for ARM-based compute modules) and other well-known software infrastructures, M2DC will also feature optimized runtime software implementations when needed for improving the efficiency of the system towards application domains such as cloud computing, big data analytics and HPC applications.

The results of these three pillars will be combined to produce TCO (Total Cost of Ownership)-optimized appliances, deployed in a real data centre environment and seamlessly interacting with existing infrastructure to run real-life applications.

The list of technical objectives consists of:

• [O1] Development of the new class of appliances with minimised Total Cost of Ownership. M2DC will deliver appliances that will provide significantly lower costs in terms of combined CAPEX (Capital Expenditures) and OPEX (Operating Expense) for data centres. Target: Costs for application users within the project should be reduced by a factor of 2.
• [O2] Reduction of energy consumed by the appliances. M2DC Appliances will use next-generation microserver modules based on latest low-power heterogeneous SoCs coming from server or mobile and embedded markets. Their smart integration in a heterogeneous system including resources such as x86, GPU or FPGA-based architectures and built-in efficiency enhancements using reconfigurable chips will deliver unprecedented efficiency while meeting quality of service demands of relevant workloads. Target: Energy efficiency improvements should reach up to 10-100 times for selected use cases and applications compared to typical servers used by  the users in 2013.
• [O3] Development of built-in dependability functions. M2DC appliances will include dependability features by design. This objective will be achieved by integration of built-in hardware-supported monitoring and management features, fast and efficient techniques for pattern matching in order to prevent failures and security threats, as well as solutions for easy maintenance and integration with external and existing systems. These functions will be implemented and integrated into turnkey appliances through the use of specific low level middleware and reconfigurable hardware. Target: M2DC appliances will contain pre-integrated security and failure detection mechanisms.
• [O4] Versatility, customisation and scalability thanks to fully software-defined design. Appliances will allow dynamic customisation to specific application workloads including their built-in functions and communication, power and thermal management (power & temperature caps, energy budgets) features. First, M2DC will extensively use the selection methodology from the FP7 FiPS project to select (possibly heterogeneous) hardware in the design and configuration phase. Second, it will also enable on-the-fly hardware customisation and adaptation by dynamic reconfiguration of FPGAs and communication facilities, include all needed software pre-installed and integrated, and add functionality to scale and integrate within data centres. Target: Customization of the M2DC appliance to a given application will take at least twice less time than providing tailored solutions using today’s heterogeneous servers.
• [O5] Intelligent power management. M2DC will achieve additional significant improvements in efficiency by integrating software and hardware techniques at various levels including: hardware-assisted system efficiency enhancements, taking advantage of novel fine-grained virtualisation techniques, optimal mapping to heterogeneous resources, optimised thermal management, and interfaces to data centre management systems. Target: Energy efficiency improved by a factor of 2-10 only using these techniques.
• [O6] Easy and powerful integration with existing data centre ecosystems. M2DC appliances will enable dynamic management and scalability by easy adaptation to an existing data centre eco-system as well as cloud and HPC systems. This integration will provide flexibility; easy deployment, provisioning, and maintenance; precise and dynamic control of power and temperature at data centre level. Hence, it will allow reducing power and cost also within a bigger ecosystem – the whole data centre. Target: Particular M2DC appliances will be integrated into BEYOND and PSNC data centres using parts of existing infrastructure and proposed to early customers and researchers.
• [O7] Enabling easy adoption of cutting-edge hardware technologies for selected applications. In addition to the latest ARM-based server and embedded processors at the latest technology nodes, M2DC will apply cutting-edge solutions such as reconfigurable hardware and HMC memory based on 3D-stacking, optical interconnects, and new manycore SoCs, to improve efficiency and limit power/temperature of the system. To enable easy adoption, M2DC will leverage existing software and use these technologies to provide built-in system enhancements functions and will benchmark relevant applications to prove the advantages of the approach. Target: Adoption and evaluation of cutting-edge technologies for at least one real industrial and scientific application.
• [O8] Demonstration of appliances for a broad set of relevant applications. One of the main challenges when designing a heterogeneous low-power system using computing resources such as ARM-based SoCs and accelerators is to identify the proper configuration of the system with respect to applications. Another challenge is to port the applications to the configured systems and to deliver sufficient QoS to the users. M2DC will strongly focus on software aspects pursuing a concept of hardware-software co-design. M2DC will ensure reliable and efficient execution of applications and other software required to implement M2DC use-cases. It will support a wide spectrum of growing classes of important applications, e.g., multimedia processing, online big data analytics for IoT and/or CyberPhysical systems, cloud computing, and HPC applications. Target: Demonstrate M2DC appliance use and efficiency for diverse applications in real environments (photo processing, IoT data analytics, cloud PaaS, and HPC tools /applications).