In the general framework of DC4Cities project, WP6 is devoted to experimentation and validation of the research outcomes of the technical work packages. The approach of these tests will follow a spiral lifecycle model: two phases of pilot tests will iteratively refine and add complexity to the workloads (real or simulated) applied to technical components in the different scenarios.
- Barcelona Trial: Two trial sites (CSUC and IMI) focused to IaaS and batch processing with different hardware profiles.
- Trento Trial: APSS site (CN) focused to application layer and batch processing.
- Milan Trial: HP Site (Laboratory) focused to web application layer over ultra-efficient bare metal hardware in a multisource energy environment.
The main Goals achieved during Phase1 are:
- Initial measurement & verification
- Analysis and definition of the different use cases by trial. (Green-SLA’s, working modes…)
- Deployment of Trial ICT infrastructure and monitoring systems.
- Deployment of cloud middle-ware orchestrator’s.
- Development of initial use case components to interact with EASC.
- Success trials execution.
Ren% = Renewables in percent. Defines the amount of renewables in comparison to the brown energy.
APC = Adaptability Power Curve. Measures the degree how much the data centre has adapted its energy consumption to a planned energy curve.
DCA = Data Centre Adaptability. Determines how much the DC energy profile has shifted from a baseline energy consumption after the implementation for flexibility mechanisms.
Infrastructure and Architecture
- Based on a real scenario of the APSS data centre in Trento.
- Generation/Processing of medical reports
- The real environment is simulated by servers in a smaller scale (one laptop and three IBM rack servers).
- Energy ecosystem contains no renewables.
- Two setups, one where the servers are all running and one with the laptop as caching utility.
- SLA: More than 800 reports per day.
- 6 different working modes.
In regard to Trento trial, which are based only in a time-shifting application use case, two distinguished behaviors have been observed. On the one hand, an optimization approach that concentrates all the work to do in those hours in which a higher percentage of RES has been analyzed in depth: this optimization is the prevalent and follows the availability of renewable energies in the grid, as can be drawn comparing RenPercent KPI in baseline and trial scenarios. In terms of flexibility, results have been very positive where the results obtained by the Trento Trial demonstrate the ability of DC4Cities to adapt the work of an application to the energy conditions. The lack of improvement in the first configuration was because of the reason, that the servers were switched never in idle modes, also not in time periods when no work was to do.
Infrastructure and Architecture
- HP-Live web application (3 m. requests per day).
- SLA: availability of e-learning platform for request handling.
- 8 PV plants on roof [550W – 1450W].
- 6 different working modes (web front-ends turned on/off).
- Front- and backend server with different host activities.
In regard to HP experiment that is using a dual power source (PVs and grid) and it is focused on infrastructure usage optimization, the results were in line with the expectations for the increase in RES usage. Since the application always needs to be in execution (and therefore no time shift is possible), only the amount of computing resources dedicated to the service can be tuned based primarily on workload and SLA (that should not be violated) and renewable energy availability.
Different PV configurations (using more or less PVs) have been compared considering the various profile days looking at the renewable usage metrics, as well as the total energy savings and application efficiency – where DC4Cities showed very significant improvements against the baseline. Looking at these results, the intrinsic nature of the HP experiment trial application is expected to provide good results in a federated scenario, further improving these initial RES metrics.
For this case we talk about video conversion services. There are many institutional repositories and portals that preserve different kind of digital media, amongst them high quality videos. These videos are available and have to be broadcasted, but in order to do so, a conversion is needed because of the huge size of videos, that are not possible to provide in their original format due to some networking infrastructure limitation.
Infrastructure and Architecture
- Two sites (CSUC and IMI) with a Opennebula cloud architecture deployed.
- Video transcoding service.
- SLA: videos in different resolutions and formats for broadcasting.
- Using different working plans for further knowledge.
- 5 days in different working modes.
- Wide range of results.
- Eager vs. aggressive plan.
In regard to Barcelona trials, which are based only in a time-shifting application use case, two distinguished behaviors have been observed. On the one hand, an optimization approach that concentrates all the work to do in those hours in which a higher percentage of RES has been analyzed in depth: this optimization is the prevalent and follows the availability of renewable energies in the grid, as can be drawn comparing RenPercent KPI in baseline and trial scenarios. For profile days with a high difference of RES availability between day and night, improvements next to 25% in percentage terms have been achieved. In terms of flexibility, results have been very positive: both in Trento and Barcelona trials the flexibility achieved in comparison to baseline scenario is high, as can be observed from the results of DCA.
The main objectives of this first set of trials have been accomplished. On the one hand, DC4Cities control system has been successfully installed in all trial environments and different executions have been run in an energy adaptive mode, achieving a higher flexibility in comparison to baseline scenario. On the other hand, results obtained and analyzed through the M&V methodology allowed to detect improvement points that have been collected as feedback to be used for the different technical WPs for refining DC4Cities components. It is worth mentioning that this feedback has been continuous during the trials and several minor changes have been already implemented.
Moreover, all three test environments have been able to adapt their consumption to the demand order produced by DC4Cities, achieving almost an APC equal to 1, which would mean that both curves has the same shape.
For more information, please contact us under: firstname.lastname@example.org