One of the new energy optimization directions addressed by the research scope of the All4Green project is improved workload control through collaborations between Data Centres (DCs), in the form of Data Centre Federations. Several Data Centres that share the same interest to collaborate towards reducing their energy footprint while maintaining their quality of service, can federate and collaborate to exchange workload among them. This collaboration is started in order to meet short time power reduction requests from energy providers and develop energy consumption plans for the short to medium term. These plans would capitalize on the price and quality attributes of the energy mix offered by different energy providers, as well as on the flexibility expressed in the Green SLAs which end-users have with their data centers.
A Data Centre federation is a set of separately managed Data Centres, with a collaborative attitude, which emerge as decentralized energy ecosystems, based on market mechanisms (Green SLAs plus dynamic negotiations) for controlling the energy consumption and CO2 emissions.
Data Centre nodes can support each other in their joint collaboration goals within the eco-system by applying not only technical innovation measures (such as, lower consumption and re-scheduling of non-critical tasks, or turning off of non-critical and/or unused servers and power and cooling systems) but also by applying better workload planning through several measures:
- DC Workload Optimization Measures: when a DC server is not servicing requests on its virtual machine instances, it is either turned off or hibernated, within the limits allowed by the Green SLAs applicable for the DC site
- DC Workload consolidation: moving all related tasks on the machines which are most frequently and most heavily used. This can be done on demand, in which case through signaling and negotiations between the DCs in the DC federation, or it can be performed in combination with Server Consolidation and with Virtual Machine Migration and/or Replication.
- DC Workload Control through re-scheduling of tasks is based on Green SLAs and negotiations between DC and End User agents: workload shifting through re-scheduling: tasks are re-planned with minimal impact on green SLAs with End Users; and workload migration: workload is shifted between the individual DCs and/or redistributed across the servers of a DC federation, with the goal to minimize the overall energy consumption at server site level.
Optimizing these collaboration measures in an integrated way in the Data Centre Federation ecosystem leads to a situation in which not only the energy provider alone is responsible for detecting and reacting to peak energy demands or a sudden supply of renewable energy, but the Data Centres play an increasing role, through collaboration based on market mechanisms and through agreements with the End User, in controlling the energy consumption and emissions, and the Quality of Service.