System and method for controlling energy usage in a server
Inventors
Vaidyanathan, Kalyanaraman • Gross, Kenneth C. • Belanger, David • Coskun, Ayse Kivilcim
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10101784-B2
Publication Date
2018-10-16
Expiration Date
Abstract
A system for controlling energy usage in a server having a processor, where the system includes a memory for storing energy cost information, and a controller for determining a transaction rate for the processor. The controller is also for determining a cumulative of energy expended by the server based on the determined transaction rate for each of a number of available power level states (P-states) for operation of the processor, and for selecting one of the available P-states for operation of the processor based on the determined cumulative energy expended and the stored energy cost information.
Core Innovation
The invention relates to a system for controlling energy usage in a server that includes at least one microprocessor and a plurality of power level operating states (P-states). For each particular P-state supported by the at least one microprocessor, the system performs a particular workload while operating at the particular P-state, and determines a cumulative of energy expended for performing the particular workload while operating at the particular P-state.
The system then determines a lowest cost P-state for operation of the at least one microprocessor based on the determined cumulative energy expended for each of the plurality of P-states and stored energy cost information. The stored energy cost information includes electricity energy-cost information that corresponds to how energy cost is evaluated, including instantaneous energy cost, fixed energy cost, or variable energy cost.
The approach emphasizes correcting prior reliance on average power curves by optimizing an integrated consumed-energy cost rather than power alone, using stored energy cost information together with the cumulative energy expended at each P-state. The selection of the lowest cost P-state is optionally constrained using a user selected target computational throughput, and a controller performs the P-state selection and then operates the at least one microprocessor in the determined lowest cost P-state.
Claims Coverage
The independent claims are directed to a system, a method, and a non-transitory storage medium. Across the independent claims, the core coverage includes iterating over supported P-states, determining cumulative energy expended for a particular workload at each P-state, and determining and operating in a lowest cost P-state based on stored energy cost information. The independent claim set explicitly identifies three ways the energy cost information can be represented: instantaneous, fixed, and variable.
Iterating over P-states for workload execution and cumulative energy determination
for each particular power level operating state (P-state) of a plurality of power level operating states (P-states) supported by the at least one microprocessor: perform a particular workload by the at least one microprocessor, while operating at the particular P-state; determine a cumulative of energy expended by the at least one microprocessor based for performing the particular workload while operating at the particular P-state.
Lowest cost P-state selection based on cumulative energy and stored energy cost information
determine a lowest cost P-state of the plurality of P-states for operation of the at least one microprocessor, based on the determined cumulative energy expended for each of the plurality of P-states, and the stored energy cost information; and based on the determination of the lowest cost P-state, operate the at least one microprocessor in the determined lowest cost P-state of the plurality of P-states.
User-selected throughput constrained P-state choice
determining, by the controller, a lowest cost P-state of the plurality of P-states, based on the determined cumulative energy expended for each of the plurality of P-states, and energy cost information; based on the determination of the lowest cost P-state, operating the at least one microprocessor in the determined lowest cost P-state of the plurality of P-states.
Energy cost information as instantaneous energy cost
the stored energy cost information includes an instantaneous energy cost.
Energy cost information as fixed energy cost
the stored energy cost information includes a fixed energy cost.
Energy cost information as variable energy cost
the stored energy cost information includes variable energy cost.
Non-transitory storage medium executing controller instructions
A non-transitory storage medium having stored computer executable instructions for controlling energy usage in a server comprising at least one microprocessor, the instructions for execution by a controller to: for each particular power level operating state (P-state) of a plurality of power level operating states (P-states) supported by the at least one microprocessor: perform a particular workload; determine a cumulative of energy expended; determine a lowest cost P-state based on the determined cumulative energy expended and energy cost information; and based on the determination of the lowest cost P-state, operate the at least one microprocessor in the determined lowest cost P-state.
Across the independent claims, the invention selects a lowest cost P-state by performing a workload at each supported P-state, determining a cumulative of energy expended for the workload at each P-state, and using stored energy cost information to determine and operate in the lowest cost P-state. Dependent coverage explicitly defines energy cost information as instantaneous, fixed, or variable energy cost, and further narrows selection using a user selected target computational throughput. The independent claim set also includes a non-transitory storage medium that executes controller instructions to perform the same P-state selection logic.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Server energy usage control in a datacenter context with energy cost based on electricity energy-cost information, including variable electricity rates.
Operating a microprocessor in a selected lowest cost P-state based on measured/derived transaction rate and cumulative energy expended for a workload.
Interested in licensing this patent?