RYAX provides power and energy monitoring and adapted scheduling techniques for the edge-cloud continuum and Serverless based workflows
Power Management
Infrastructures for digital technologies consume around 8-10% of the total energy consumption and account for 2-4% of the global greenhouse gas emissions [EUGD].
Power monitoring on IT infrastructures is needed to control the usage of the most important resource: that of power, which internally in the systems, should be treated as first class citizen.
Reasons for Power and Energy Monitoring
Power/energy attribution usage: Fine-grained monitoring can provide details on how the consumed energy of IT systems is utilized among the users, projects, departments, etc
Carbon emissions estimation: This can be done through power monitoring and combination with other factors.
User incentive: Power and energy details along with CO2 emissions related to the task execution can provide interesting insights to users which can motivate them to optimize their executions energy consumption and footprint.
Adapted accounting & billing: Usage of power, energy and footprint can be charged to the users based on particular policies and formulas.
Automated optimizations: Analysis of power and energy measurements can help us propose various system-level optimizations for power/energy efficient scheduling and placement and contribute in decreasing the PUE (Power Usage Effectiveness: metric which measures the total amount of power that a data center consumes)
Techniques for Power and Energy Monitoring
Different techniques exist to enable power and energy monitoring on IT systems
- External Power/Watt meters which can be installed in the data center to provide accurate and global measurements.
- Internal node/chip counters such as RAPL (Intel CPUs) and NVML (Nvidia GPUs) are available and can provide power/energy measurements with good accuracy and high sampling.
Power and Energy Efficient scheduling
Various methods exist and can be implemented to provide automated power and energy consumption optimizations on the system-level for the data center or the continuum
Power-capping: Manage and control the instantaneous power peaks of the data center through automated adaptation of the execution of tasks.
Energy-aware scheduling: Consider the energy consumption of tasks and workflows running on the data center and adapt their execution in space and time in such a way in order to respect particular energy budgets.
Thermal-aware placement: Consider the temperature of the different nodes of the data center and schedule the tasks on nodes that have lower temperature in order to avoid hot spots in the data-center
Carbon-footprint aware scheduling: Control the placement of the tasks and workflows based on the type or cost of the electricity used on the data-centers; Favor data-centers or nodes that use Renewable Energy Sources.
RYAX and Power and Energy monitoring and scheduling
Within R&D European project PHYSICS "Optimized Hybrid Space-Time Continuum in FaaS",
RYAX provides power and energy monitoring and adapted scheduling techniques for the edge-cloud continuum and Serverless based workflows.
Usage of high level power/energy aggregators for the data center enable the power/energy monitoring of multiple nodes of the data center such as Kepler and Scaphandre
Adapted multi-objective energy-aware scheduling techniques based on 2-level scheduling method for serverless-based continuum
PHYSICS : Optimized Hybrid Space-Time Continuum in FaaS, https://physics-faas.eu/