Search

EP-4468114-B1 - DATA CENTER SYSTEM, INTER-BASE WORKLOAD CONTROL METHOD, AND INTER-BASE WORKLOAD CONTROL SYSTEM

EP4468114B1EP 4468114 B1EP4468114 B1EP 4468114B1EP-4468114-B1

Inventors

  • OKAMURA, TAKU
  • KANEKO, SATOSHI

Dates

Publication Date
20260506
Application Date
20240305

Claims (8)

  1. A data center system (S) comprising: a plurality of bases (1000) provided with a calculation resource (3000, 4000) and a renewable energy power supply (7000) to execute a workload (WL); and an inter-base workload control system (10000) to control migration of the workload (WL) between the plurality of bases (1000) and within a same base, wherein a processor (19000) of the inter-base workload control system (10000) is configured to manage: a positive excess power amount obtained by subtracting a predicted amount of power supply of the renewable energy power supply (7000) from a predicted amount of power consumption of the calculation resource (3000, 4000) associated with execution of the workload (WL) in a future time range at the bases; spatial migratable time range information (12345) on a spatial migratable time range where spatial migration is possible and the workload (WL) scheduled to be executed in the future time range at the bases can be migrated to another base; spatial non-migratable time range information (12350) on a non-migratable time range where it is not possible to migrate a workload (WL) scheduled to be executed in the future time range to another base; temporal migratable time range information (12335) on temporal migration whereby the workload (WL) scheduled to be executed in the future time range at the bases is delayed in execution within the same base to be migrated to another time range; and a predicted amount of power consumption (12310) which is a predicted amount of power consumption of the calculation resource (3000, 4000) associated with execution of the workload (WL) scheduled to be executed in the future time range at the bases, to determine a power adjustment amount indicating the predicted amount of power consumption (12310) that moves through the spatial migration and the temporal migration of the workload (WL) scheduled to be executed in the future time range at the bases based on the spatial migratable time range information (12345), the spatial non-migratable time range information (12350), the temporal migratable time range information (12335) and the predicted amount of power consumption (12310) so that a sum total of the excess power amounts in the future time range at the plurality of bases (1000) becomes smaller, and to execute the spatial migration and the temporal migration of the workload (WL) based on the power adjustment amount.
  2. The data center system (S) according to claim 1, wherein when the amount of excess power exists in the future time range at the bases, the processor (19000) is configured to determine the power adjustment amount that moves through the spatial migration of the workload (WL) scheduled to be executed in a specific time range at a specific base where the amount of excess power exists so that the sum total becomes smaller, and after the power adjustment amount is determined, if the amount of excess power further exists in the specific time range at the specific base, the processor (19000) is configured to determine a first power adjustment amount indicating the predicted amount of power consumption (12310) that moves through the temporal migration of the first workload scheduled to be executed in the specific time range at the specific base so that the sum total further becomes smaller, wherein the processor (19000) is configured to determine the first power adjustment amount for temporal migration of the first workload and to execute the temporal migration based on the determined first power adjustment amount.
  3. The data center system (S) according to claim 2, wherein the processor (19000) is configured to determine whether or not there is a second workload scheduled to be executed, the spatial migration of which is possible in the time range next to the specific time range, and when there is the second workload, the spatial migration of which is possible in the time range next to the specific time range, the processor (19000) is configured to determine a second power adjustment amount indicating the predicted amount of power consumption (12310) that moves through the spatial migration of the second workload as the first power adjustment amount.
  4. The data center system (S) according to claim 3, wherein when the amount of excess power in the specific time range is equal to or smaller than the second power adjustment amount, the processor (19000) is configured to determine the first power adjustment amount by which to delay, wherein due to the delay by the first power adjustment amount, part or all of the amount of excess power in the specific time range moves to the next time range up to the amount of excess power.
  5. The data center system (S) according to claim 3, wherein when the amount of excess power in the specific time range exceeds the second power adjustment amount, the processor (19000) is configured to determine the first power adjustment amount by which to delay, wherein due to the delay by the first power adjustment amount, part of the amount of excess power in the specific time range moves to the next time range up to the second power adjustment amount.
  6. The data center system (S) according to claim 2, wherein the data center system (S) is configured to determine the workload (WL) to be a spatial migration target in the future time range where the spatial migration is possible based on the power adjustment amount, and to execute the spatial migration for the determined workload (WL).
  7. The data center system (S) according to claim 2, wherein the data center system (S) is configured to determine the workload (WL) as a temporal migration target in the future time range where the spatial migration is not possible based on the first power adjustment amount, and to execute the temporal migration for the determined workload (WL).
  8. An inter-base workload control method executed by an inter-base workload control system (10000), the inter-base workload control system (10000) comprising a plurality of bases (1000) comprising a calculation resource (3000, 4000) and a renewable energy power supply (7000) to execute a workload to control migration of the workload (WL) between the plurality of bases (1000) and within a same base, the method comprising steps executed by a processor (19000) of the inter-base workload control system (10000) managing: a positive excess power amount obtained by subtracting a predicted amount of power supply of the renewable energy power supply (7000) from a predicted amount of power consumption of the calculation resource (3000, 4000) associated with execution of the workload (WL) in a future time range at the base, spatial migratable time range information (12345) on a spatial migratable time range where spatial migration of migrating the workload (WL) scheduled to be executed in the future time range at the base to another base is possible, spatial non-migratable time range information (12350) on a non-migratable time range where it is not possible to migrate a workload (WL) scheduled to be executed in the future time range to another base, temporal migratable time range information (12335) on temporal migration of delaying execution of the workload (WL) scheduled to be executed in the future time range at the base within a same base and migrating the workload (WL) to another time range and a predicted amount of power consumption (12310) which is a predicted amount of power consumption of the calculation resource (3000, 4000) associated with execution of the workload (WL) scheduled to be executed in the future time range at the base, determining a power adjustment amount indicating the predicted amount of power consumption (12310) that moves through the spatial migration and the temporal migration of the workload (WL) scheduled to be executed in the future time range at the bases based on the spatial migratable time range information (12345), the spatial non-migratable time range information (12350), the temporal migratable time range information (12335) and the predicted amount of power consumption (12310) so that a sum total of the excess power amounts in the future time range at the plurality of bases (1000) becomes smaller; and executing the spatial migration and the temporal migration of the workload (WL) based on the power adjustment amount.

Description

CROSS-REFERENCE TO RELATED APPLICATION This application is based upon and claims the benefit of priority of the prior Japanese Patent Application No. 2023-86288, filed on May 25, 2023. BACKGROUND The present invention relates to a data center system, a DC inter-base workload control method and a DC-inter-base workload control system. Description of the Related Art Attention has been attracted to so-called decarbonization, which aims to move away from fossil fuels so as to prevent emissions of greenhouse gases such as carbon dioxide, which causes global warming. In this respect, many information processing apparatuses and communication equipment to execute predetermined processing loads (workloads) are set in data centers (DCs), and operating these apparatuses and equipment requires a large amount of electric power. Therefore, attempts are being made to achieve decarbonization by providing such electric power from renewable energy sources. In this case, while it is important to keep a ratio of renewable energy to power consumption (renewable energy utilization ratio), it is preferable to keep this renewable energy utilization ratio in order to achieve cleaner DCs with finer time granularity (e.g., hourly rather than daily). Another problem is that supplies of renewable energy depend on factors such as weather, and so energy supplies are quite unstable, and as the dependency on renewable energy increases, power providers need to be provided with more coordinating power (power necessary to match demand and supply of electricity) so that the power providers perform stable power supplies, and this constitutes a burden on the power providers. Therefore, there are attempts to control power consumed at DCs by temporally controlling workloads executed at DCs or by spatially controlling the workloads between a plurality of DC bases (collection of DCs within the same area) to control power consumed at DCs and thereby improve the renewable energy utilization ratio or utilize the DCs as coordinating power for the power system. Here, "temporally controlling a workload" refers to such control that a workload is temporarily delayed in execution within the same DC base. In this way, by delaying many workloads from a time range where renewable energy supply is insufficient to a time range where renewable energy supply is surplus, it is possible to increase the renewable energy utilization ratio. On the other hand, "spatially controlling a workload" refers to such control that a location where a workload is executed is moved from a certain DC base to another DC base. In this way, as in the case of temporal control, by moving the workload from a DC base where renewable energy supply is insufficient to another DC base where renewable energy supply is abundant, it is possible to increase the renewable energy utilization ratio. Generally, workloads executed at DC bases can be broadly divided into two types: interactive workload and batch job. An interactive workload is one such as a Web application that needs to be processed in real time and is basically always in a running condition. A batch job is one such as image processing or machine learning training process that does not necessarily need to be processed immediately, and may be executed by a predetermined time and processing thereof can be delayed within such a range. Although the interactive workload here can be spatially controlled, processing of the interactive workload cannot be delayed, and therefore the interactive workload is not suitable for temporal control. On the other hand, processing of a batch job can be delayed, and so it can be temporally controlled. Thus, for clarity, the former will be referred to as a "spatial-migration-oriented workload" and the latter will be referred to as a "temporal-migration-oriented workload." Japanese Patent Laid-Open No. 2021-189845 discloses a technology to efficiently utilize renewable energy generated at DC bases by migrating workload execution process from a DC base where renewable energy supply is predicted to be insufficient to a DC base where renewable energy supply is predicted to be surplus. Additionally, US 2014 / 365 402 A1 discloses a data center system that utilizes multiple power sources, including renewable energy, to manage computational tasks efficiently. It features a management system that coordinates both temporal and spatial movement of data components between data centers based on power availability and cost considerations. Using the technology in Japanese Patent Laid-Open No. 2021-189845 makes it possible to perform spatial control of workloads according to renewable energy supply at the DC bases and adjust the ratio of workloads executed at the DC bases. This allows the renewable energy utilization ratio to be improved at the DC bases. However, workloads executed at the DC bases cannot always migrate between the DC bases freely. When workloads migrate between the DC bases, downtimes may be produced on no s