Dynamic peak power limiting to processing nodes in an information handling system
First Claim
1. A computer-implemented method to dynamically limit peak power consumption in processing nodes of an information handling system (IHS), the method comprising:
- receiving, at a power management micro-controller (PMM), a plurality of processing node-level power-usage and workload data from a plurality of node controllers, including a current power consumption and a current workload, for each processing node within the IHS, wherein each of the processing nodes has a complex programmable logic device (CPLD) and a board management controller (BMC), the CPLD coupled to the PMM via an I2C bus, and the BMC coupled to the PMM via an Ethernet cable, the PMM providing control and/or management signals to the processing nodes via the I2C bus and one or more select wires within the Ethernet cable;
identifying a total available system power of the IHS including a peak power output capacity and a sustained output power capacity;
receiving a processing node priority data from a management controller (MC) that is communicatively coupled to each of the plurality of node controllers of the processing nodes;
storing the priority data to PMM memory;
determining, by the micro-controller, at least one node peak power threshold based on the power-usage and workload data for each of the processing nodes and in part based on the processing node priority data;
determining, by the micro-controller, at least one node average power threshold based on the power-usage and workload data for each of the processing nodes and in part based on the processing node priority data;
transmitting the node peak power threshold and the node average power threshold to the node controllers of each of the processing nodes;
triggering the node controllers to determine and set a device peak power limit for each of a plurality of respective variable performance devices located within or associated with each of the processing nodes based on the node peak power threshold;
triggering each of the plurality of node controllers to determine and set a respective device average power limit for each of the plurality of respective variable performance devices associated with each of the processing nodes based on a respective node average power threshold;
wherein each of the plurality of respective variable performance devices concurrently adjusts a corresponding device performance metric, from among an operating frequency and a data throughput, that applies to a respective one of the variable performance devices, based on the average power limit and a peak power limit of the respective device;
wherein each of the respective variable performance level devices dynamically adjusts, based on the device peak power limit, the corresponding device performance metric, in order for a current peak power level to remain within or less than the device peak power limit;
storing the at least one node peak power threshold and at least one average power threshold in node controller memory;
determining if a current device peak power consumption is approaching the device peak power limit for at least one of the devices; and
in response to determining that the current device peak power consumption is approaching the device peak power limit, the corresponding device estimating its current peak power level based on the workload data and adjusting the at least one of the operating frequency and the data throughput of the corresponding device such that the current peak power level remains less than the device peak power limit.
22 Assignments
0 Petitions
Accused Products
Abstract
A computer-implemented method dynamically limits peak power consumption in processing nodes of an IHS. A power management micro-controller receives processing node-level power-usage and workload data from several node controllers, including current power consumption and a current workload, for each processing node within the IHS. A total available system power of the IHS is identified including a peak power output capacity and a sustained output power capacity. At least one node peak power threshold is determined based on the power-usage and workload data for each of the processing nodes. The node controllers are triggered to determine and set a device peak power limit for each of several variable performance devices within each of the processing nodes based on the node peak power threshold, wherein each of the variable performance devices dynamically adjusts a value of a corresponding device performance metric based on the device peak power limit.
37 Citations
19 Claims
-
1. A computer-implemented method to dynamically limit peak power consumption in processing nodes of an information handling system (IHS), the method comprising:
-
receiving, at a power management micro-controller (PMM), a plurality of processing node-level power-usage and workload data from a plurality of node controllers, including a current power consumption and a current workload, for each processing node within the IHS, wherein each of the processing nodes has a complex programmable logic device (CPLD) and a board management controller (BMC), the CPLD coupled to the PMM via an I2C bus, and the BMC coupled to the PMM via an Ethernet cable, the PMM providing control and/or management signals to the processing nodes via the I2C bus and one or more select wires within the Ethernet cable; identifying a total available system power of the IHS including a peak power output capacity and a sustained output power capacity; receiving a processing node priority data from a management controller (MC) that is communicatively coupled to each of the plurality of node controllers of the processing nodes; storing the priority data to PMM memory; determining, by the micro-controller, at least one node peak power threshold based on the power-usage and workload data for each of the processing nodes and in part based on the processing node priority data; determining, by the micro-controller, at least one node average power threshold based on the power-usage and workload data for each of the processing nodes and in part based on the processing node priority data; transmitting the node peak power threshold and the node average power threshold to the node controllers of each of the processing nodes; triggering the node controllers to determine and set a device peak power limit for each of a plurality of respective variable performance devices located within or associated with each of the processing nodes based on the node peak power threshold; triggering each of the plurality of node controllers to determine and set a respective device average power limit for each of the plurality of respective variable performance devices associated with each of the processing nodes based on a respective node average power threshold; wherein each of the plurality of respective variable performance devices concurrently adjusts a corresponding device performance metric, from among an operating frequency and a data throughput, that applies to a respective one of the variable performance devices, based on the average power limit and a peak power limit of the respective device; wherein each of the respective variable performance level devices dynamically adjusts, based on the device peak power limit, the corresponding device performance metric, in order for a current peak power level to remain within or less than the device peak power limit; storing the at least one node peak power threshold and at least one average power threshold in node controller memory; determining if a current device peak power consumption is approaching the device peak power limit for at least one of the devices; and in response to determining that the current device peak power consumption is approaching the device peak power limit, the corresponding device estimating its current peak power level based on the workload data and adjusting the at least one of the operating frequency and the data throughput of the corresponding device such that the current peak power level remains less than the device peak power limit. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12)
-
-
13. An information handling system (IHS) comprising:
-
one or more processing nodes, each of the processing nodes having at least one node controller, the node controllers each controlling node-level operations of the processing nodes within a corresponding node, the processing nodes each having at least one central processing unit (CPU) communicatively coupled to the node controller; one or more variable performance level devices communicatively coupled to the node controller and having a performance level based on a value of a corresponding device performance metric; a power management module having a power management micro-controller and a memory coupled to the power management micro-controller, the power management module communicatively coupled to the processing nodes, wherein each of the processing nodes has a complex programmable logic device (CPLD) and a board management controller (BMC), the CPLD coupled to the PMM via an I2C bus, and the BMC coupled to the PMM via an Ethernet cable, where the PMM provides control and/or management signals to the processing nodes via the I2C bus and one or more select wires within the Ethernet cable; a power subsystem communicatively coupled to the power management micro-controller and providing power distribution to the IHS, the power subsystem including at least one power supply unit (PSU); and the power management micro-controller having firmware executing thereon to enable dynamic peak power limiting to the processing nodes within the IHS, wherein the firmware configures the power management micro-controller to; receive a plurality of processing node-level power-usage and workload data from the node controllers, including a current power consumption and a current workload, for each processing node within the IHS; identify a total available system power of the IHS including a peak power output capacity and a sustained output power capacity; receiving a processing node priority data from a management controller (MC) that is communicatively couples to the processing nodes; storing the priority data to PMM memory; determine at least one node peak power threshold based on the power-usage and workload data for each of the processing nodes and in part based on the processing node priority data; determine at least one node average power threshold based on the power-usage and workload data for each of the processing nodes and in part based on the processing node priority data; store the at least one node peak power threshold and at least one average power threshold in node controller memory; transmit the node peak power threshold and the node average power threshold to the node controllers of each of the processing nodes; and trigger the node controllers to determine and set a device peak power limit for each of the at least one variable performance level device within each of the processing nodes based on the node peak power threshold; trigger each of the plurality of node controllers to determine and set a respective device average power limit for each of the plurality of respective variable performance devices associated with each of the processing nodes based on a respective node average power threshold; wherein each of the plurality of respective variable performance devices concurrently adjusts a corresponding device performance metric, from among an operating frequency and a data throughput, that applies to a respective one of the variable performance devices, based on the average power limit and a peak power limit of the respective device; wherein each of the at least one variable performance level device dynamically adjusts the corresponding device performance metric based on the device peak power limit, in order for a current peak power level to remain within or less than the device peak power limit; determine if a current device peak power consumption is approaching the device peak power limit for at least one of the devices; and in response to determining that the current device peak power consumption is approaching the device peak power limit, the corresponding device estimates a current peak power level based on the workload data and adjust the at least one of the operating frequency and the data throughput of the corresponding device such that the current peak power level remains less than the device peak power limit. - View Dependent Claims (14, 15, 16, 17, 18, 19)
-
Specification