Back to results
US 9,557,792 B1Granted
Datacenter power management optimizations
AssigneeAmazon Technologies, Inc.
InventorsNachiketh Rao Potlapally, James R. Hamilton
Life of this patent
Estimating the term…
Abstract
Methods and apparatus for datacenter power management optimization are disclosed. Metrics, including workload data, thermal measurements and the like are collected from numerous endpoints within a datacenter. System profiles of a plurality of servers, and application workload profiles for various workloads, are stored. Based on analysis of collected metrics, power optimization operations comprising either workload scheduling operations, power configuration change operations, or both, are initiated.
Claims24
A system, comprising:
- one or more computing devices configured to:
- collect, from a plurality of power management endpoints at a data center of a provider network, respective sets of metrics, wherein the data center comprises a plurality of servers, wherein the plurality of power management endpoints includes at least one power distribution unit (PDU) endpoint, at least one rack endpoint, and at least one server endpoint, and wherein a set of metrics collected from a particular server endpoint comprises an indication of power usage at a particular server of the plurality of servers and an indication of a particular application executing at the particular server;
- store (a) respective location mappings for the plurality of servers of the data center, wherein a location mapping of the particular server comprises an indication of a rack in which the particular server is located and one or more power domains from which the particular server obtains power, (b) system profiles for the plurality of servers, and (c) workload profiles of one or more applications, including the particular application, targeted for execution at the data center, wherein each workload profile includes a respective indication of a corresponding resource consumption pattern associated with the respective application;
- determine, based at least in part on an analysis of one or more metrics collected from one or more power management endpoints of the plurality of power management endpoints, to initiate a power optimization operation;
- initiate, as part of the power optimization operation, at least one of:
- (a) a workload scheduling operation associated with the particular application, or (b) a power configuration operation associated with at least one power management endpoint of the plurality of power management endpoints;
- classify, based at least in part on their respective system profiles, the plurality of servers into a plurality of capability categories, wherein a particular server is classified in a particular capability category based at least in part on a performance level of the particular server;
- and
- schedule workload associated with the particular application on the particular server based at least in part on (a) a determination that the workload profile of the particular application corresponds to the particular capability category and (b) a determination that a power consumption profile of the particular server is in accordance with a power consumption goal.
6 more elements
A method, comprising:
- performing, by one or more computing devices:
- collecting, from a plurality of power management endpoints at a data center, respective sets of metrics;
- storing system profiles for a plurality of servers of the data center and workload profiles of one or more applications targeted for execution at the data center, wherein each workload profile includes a respective indication of a corresponding resource consumption pattern associated with the respective application;
- determining, based at least in part on an analysis of one or more metrics collected from one or more power management endpoints of the plurality of power management endpoints, to initiate a power optimization operation;
- initiating, as part of the power optimization operation, at least one of:
- (a) a workload scheduling operation associated with an application of the one or more applications or (b) a power configuration operation associated with at least one power management endpoint of the plurality of power management endpoints;
- classifying, based at least in part on their respective system profiles, the plurality of servers into a plurality of capability categories, wherein a particular server is classified in a particular capability category based at least in part on a performance level of the particular server;
- and
- scheduling workload associated with a particular application of the one or more applications on the particular server based at least in part on (a) a determination that the workload profile of the particular application corresponds to the particular capability category and (b) a determination that a power consumption profile of the particular server is in accordance with a power consumption goal.
6 more elements
A non-transitory computer-accessible storage medium storing program instructions that when executed on one or more processors:
- collect, from a plurality of power management endpoints at a data center, respective sets of metrics;
- store system profiles of a plurality of servers of the data center and workload profiles of one or more applications, wherein each workload profile includes a respective indication of a corresponding resource consumption pattern associated with the respective application;
- initiate, based at least in part on an analysis of one or more metrics collected from one or more power management endpoints of the plurality of power management endpoint, at least one of:
- (a) a workload scheduling operation associated with an application of the one or more applications or (b) a power configuration operation associated with at least one server of the plurality of servers;
- classify, based at least in part on their respective system profiles, the plurality of servers into a plurality of capability categories, wherein a particular server is classified in a particular capability category based at least in part on a performance level of the particular server;
- and
- schedule workload associated with a particular application of the one or more applications on a particular server based at least in part on (a) a determination that the workload profile of the particular application corresponds to the particular capability category and (b) a determination that a power consumption profile of the particular server is in accordance with a power consumption goal.
4 more elements
Dependencies are inferred from the claim text.