WP3: Cost-aware Workload Balancing and Consolidation

Objectives

  • Development and validation of IT work load models
  • Optimisation for energy-aware workload consolidation
  • Energy-aware geographically distributed data centre optimisation
  • Virtual workload demand forecasting
  • Automatic virtual workload control

Year 1 Achievements

  • WP3 identified the appropriate Artificial Intelligence and Machine Learning techniques that should be applied in the context of the GENiC workload allocation problem. WP3 activities also saw the development of initial prototypes that are used to test the optimisation approach, using real-world data to measure the power consumption saving achieved.

Year 2 Achievements

  • Validation of Workload models.
  • Evaluation Single data centre workload optimisation with a first version of this component being deployed in the GENiC platform.
  • Initial investigation into the problem of internet scale data centre workload optimisation.
  • Development of tools relating to computation & storage demands and workload migrator tool to distribute workload.

Year 3 Achievements

  • Energy-aware geographically distributed data centre optimisation
    The optimisation problem of allocating virtual machines to geographically distributed data centres was extended:
    - Constraints limiting migrations were enforced at a data centre level.
    - Additional variant considered where each location has on-site uncontrollable renewables whose output varies across time.
    - Extensive data related with electricity prices were collected.
    - Rather than focusing on game theoretic aspects, which we deemed to be impractical from a real-world perspective, a distributed version of the problem was defined, and the generator was adapted to handle this extension
  • Virtual workload demand forecasting and modelling were completed.
    - The forecasting technique was implemented into the GENiC workload prediction component, verified the novel prediction technique inside the GENiC platform both in the simulation-based assessment and at the real test site, and studied the applicability of the forecasting techniques outside of the workload domain, i.e. in the context of power traces and IT power consumption prediction
  • Automatic Virtual Workload Control was completed.
    - The workload control was implemented into the GENiC performance optimisation component.
    - The monitoring system has been validated through unit, integration and system tests verifying that the system is capable to support the monitoring of virtual resources across their entire lifecycle (start, stop, suspend, restart, migrate) and created monitoring templates used to incorporate within the monitoring system the measurements coming from the External Acquisition services (weather, weather forecast, grid energy prices).