Load control in a data center
AssigneeExaflop LLC
InventorsWolf-Dietrich Weber, Xiaobo Fan, Luiz Andre Barroso
Life of this patent
Estimating the term…
Abstract
A method of controlling power usage in a data center includes monitoring a power usage of a plurality of computers in the data center, generating a signal indicating that the power usage is within a threshold of a maximum power capacity, and in response to the signal, adjusting performance of at least one computer in the plurality of computers.
Claims14
A method of running multiple computing systems to maximize a number of systems in use with the amount of power available, comprising:
- identifying a first workload type that has a first request throughput and a first data processing requirement;
- identifying a second workload type that comprises an internet service and has a second request throughput and a second data processing requirement;
- identifying a third workload type that comprises an offline batch job service and has a third request throughput and a third data processing requirement, wherein the first request throughput is greater than the second and third request throughputs, and the first data processing requirement is greater than the second and third data processing requirements;
- determining a first power usage value for the first workload type, a second power usage value for the second workload type, and a third power usage value for the third workload type;
- selecting a highest power usage value from the first, second, and third power usage values;
- and
- assigning a first number of systems to run the first workload type, a second number of systems to run the second workload type and a third number of systems to run the third workload type based on the amount of power available and the highest power usage value.
A computer program product, tangibly embodied in a non-transitory computer-readable medium, comprising instructions to cause a processor to perform operations comprising:
- identifying a first workload type that has a first request throughput and a first data processing requirement;
- identifying a second workload type that comprises an internet service and has a second request throughput and a second data processing requirement;
- identifying a third workload type that comprises an offline batch job service and has a third request throughput and a third data processing requirement, wherein the first request throughput is greater than the second and third request throughputs, and the first data processing requirement is greater than the second and third data processing requirements;
- determining a first power usage value for the first workload type, a second power usage value for the second workload type, and a third power usage value for the third workload type;
- selecting a highest power usage value from the first, second, and third power usage values;
- and
- assigning a first number of systems to run the first workload type, a second number of systems to run the second workload type and a third number of systems to run the third workload type based on the amount of power available and the selected highest power usage value.
A method of safeguarding multiple computing systems in a datacenter, comprising:
- selecting a value below an actual peak power, where the actual peak power is less than about 60% of a nameplate value of one of the multiple computing systems;
- monitoring power usage of the multiple computing systems during use;
- predicting future power usage based on requested tasks over a predetermined time duration;
- determining that the future power usage exceeds the value during a portion of the predetermined time duration;
- and
- based on the determination that the future power usage exceeds the value, implementing component-level power management control functions or descheduling tasks during the portion of the predetermined time duration.
A computer program product, tangibly embodied in a non-transitory computer-readable medium, comprising instructions to cause a processor to:
- select a value below an actual peak power, where the actual peak power is less than about 60% of a nameplate value of one of the multiple computing systems;
- monitor power usage of multiple computing systems during use;
- predicting future power usage based on requested tasks over a time duration;
- determining that the future power usage exceeds the value during a portion of the time duration;
- and
- based on the determination that the future power usage exceeds the value, implementing component-level power management control functions or descheduling tasks during the portion of the time duration.
Dependencies are inferred from the claim text.