Abstract: Disclosed are a time synchronization processing method and apparatus. The method comprises: determining a time stability index of a node according to a time deviation value of the node within a preset period of time, where the time deviation value is a delay of a clock signal of the node relative to a reference clock signal within the preset period of time, and the time stability index comprises at least one of the following: a time compensation accumulated value within the preset period of time, a maximum time compensation value within the preset period of time, a time compensation average value within the preset period of time, and a time fluctuation value within the preset period of time; determining whether the time stability index of the node exceeds a preset range; and in a case that the time stability index of the node exceeds the preset range, sending a time synchronization exception alarm.

Abstract: Systems and methods are provided for distributing a system power cap amongst system equipment for efficient utilization of power cap ranges without requiring an understanding of intricacies of the system architecture. An example of the system and methods obtain power cap ranges for controllable system equipment and power cap values for non-controllable system equipment; calculate a settable power cap range for the system based on the power cap ranges and power cap values; based on a requested power cap, determine power caps for controllable system equipment from a comparison of the requested power cap against the settable power cap range; and provide the determined power caps to the system for application to each of the controllable system equipment. In various examples, the power caps for the controllable system equipment can be determined through application of one or more distribution schemes.

Abstract: A region-aware GPU power/energy regulation method comprises periodically identifying a phase of execution of an application which is currently being executed by a GPU and measuring the utilization of the GPU (e.g., memory utilization) during execution of the identified phase. The utilization may be measured during a sampling period at both high and low GPU frequencies. A frequency sensitivity parameter is then determined for the identified phase based on the measured utilization of the GPU. A selected frequency for the identified phase is then determined based on the frequency sensitivity parameter. The GPU can then be instructed to set a frequency of the GPU to the selected frequency during execution of the remainder of the identified phase.

Abstract: Automated systems and methods are provided for distributing a system power cap amongst system equipment for efficient utilization of power cap ranges without requiring an understanding of intricacies of the system architecture. Examples of the systems and methods automatically, responsive to trigger events, distribute the system power cap amongst system equipment. Another example provides for grouping of system equipment into multiple pools and distributing the system power cap to system equipment on a per-pool basis, according to a prioritized order.

Abstract: A system determines a plurality of benchmark workloads including a compute-heavy workload, a memory-heavy workload, and an input/output (I/O)-heavy workload. The system obtains, for a respective benchmark workload, a benchmark power measurement associated with a benchmark system. The benchmark power measurement indicates a target efficiency threshold for the respective benchmark workload. The system measures power characteristics for a current workload on a computing device. The power characteristics comprise current power measurements associated with the computing device, processing components of the computing device, memory components of the computing device, and I/O components of the computing device. The system identifies, based on a ratio between two of the power characteristics, a benchmark workload most closely associated with the current workload.

Abstract: Systems and methods are provided for improving statistical and machine learning drift detection models that monitor computing health of a data center environment. For example, the system can receive streams of sensor data from a plurality of sensors in a data center; clean the streams of sensor data; generate, using a machine learning (ML) model, an anomaly score and a dynamic threshold value based on the cleaned streams of sensor data; determine, using the ML model and based on the anomaly score and the dynamic threshold value, a correctness indicator for a first sensor in the plurality of sensors; and using the correctness indicator, correct the first sensor.

Abstract: A process includes communicating a coolant flow between an inlet and an outlet of a coolant subsystem that is associated with a cooling domain to remove thermal energy from a plurality of processor-based nodes of the cooling domain. The communication of the coolant flow has associated predefined parameters. The process includes regulating a temperature of the coolant flow at the outlet. In accordance with example implementations, the regulation includes determining, based on the predefined parameters, a minimum collective power consumption by the processor-based nodes to maintain a temperature of the coolant flow at the outlet at or above a minimum threshold temperature, and based on the minimum power consumption, scheduling jobs to be executed by the nodes.

Abstract: A method for optimizing operations of high-performance computing (HPC) systems includes collecting data associated with a plurality of workload performance profiling counters associated with a workload during runtime of the workload in an HPC system. Based on the collected data, the method includes using a machine-learning technique to classify the workload by determining a workload-specific fingerprint for the workload. The method includes identifying an optimization metric to optimize during running of the workload in the HPC system. The method includes determining an optimal setting for a plurality of tunable hardware execution parameters as measured against the optimization metric by varying at least a portion of the plurality of tunable hardware execution parameters.

Abstract: In an example implementation consistent with the features disclosed herein, a computer-implemented method includes monitoring power available to a plurality of nodes. Each node corresponding to an electronic component, device, or group of devices. The method includes detecting that the power available to the nodes is outside a power range within which all of the nodes can operate at a full operation level. When the power available to the node is outside the power range, allocation of the available power amongst the nodes is determined and at least some of the nodes are signaled to transition to a mode that utilizes less power.

Abstract: A region-aware power/energy regulation method comprises periodically identifying a region of an application which is currently being executed by a given processor out of a plurality of regions of the application. The method also comprises measuring the instructions per second (IPS) of the given processor during execution of the identified region. The method also comprises determining a compute-boundedness parameter of the identified region based on the measured IPS. The method also comprises determining an optimal frequency for the identified region based on the compute-boundedness parameter thereof, and instructing the given processor to set a frequency of the given processor to the optimal frequency during execution of the identified region.

Abstract: Systems and methods are provided for distributing a system power cap amongst system equipment for efficient utilization of power cap ranges without requiring an understanding of intricacies of the system architecture. An example of the system and methods obtain power cap ranges for controllable system equipment and power cap values for non-controllable system equipment; calculate a settable power cap range for the system based on the power cap ranges and power cap values; based on a requested power cap, determine power caps for controllable system equipment from a comparison of the requested power cap against the settable power cap range; and provide the determined power caps to the system for application to each of the controllable system equipment. In various examples, the power caps for the controllable system equipment can be determined through application of one or more distribution schemes.

Abstract: Automated systems and methods are provided for distributing a system power cap amongst system equipment for efficient utilization of power cap ranges without requiring an understanding of intricacies of the system architecture. Examples of the systems and methods automatically, responsive to trigger events, distribute the system power cap amongst system equipment. Another example provides for grouping of system equipment into multiple pools and distributing the system power cap to system equipment on a per-pool basis, according to a prioritized order.

Abstract: A process includes communicating a coolant flow between an inlet and an outlet of a coolant subsystem that is associated with a cooling domain to remove thermal energy from a plurality of processor-based nodes of the cooling domain. The communication of the coolant flow has associated predefined parameters. The process includes regulating a temperature of the coolant flow at the outlet. In accordance with example implementations, the regulation includes determining, based on the predefined parameters, a minimum collective power consumption by the processor-based nodes to maintain a temperature of the coolant flow at the outlet at or above a minimum threshold temperature, and based on the minimum power consumption, scheduling jobs to be executed by the nodes.

Abstract: In some examples, a controller dynamically configures a property associated with monitoring performed by an agent. The controller stores, in a repository, metadata relating to the agent. The controller receives, from the agent, first sensor data that excludes the metadata, and uses indexing information in the first sensor data to retrieve the metadata from the repository.

Abstract: A technique includes an agent executing on a plurality of nodes while a job is being concurrently executed by the plurality of nodes. The plurality of nodes is power-capped by an existing node power consumption budget. The technique includes managing power consumption of the plurality of nodes. The managing includes the agent determining a performance footprint that is associated with execution of the job; and the managing includes the agent determining a second node power consumption budget based on the performance footprint. The second node power consumption budget is different than the existing node power consumption budget. The managing includes the agent providing a power consumption request to a global power dispatcher to set a new node power consumption budget for the plurality of nodes.

Abstract: A method for optimizing operations of high-performance computing (HPC) systems includes collecting data associated with a plurality of workload performance profiling counters associated with a workload during runtime of the workload in an HPC system. Based on the collected data, the method includes using a machine-learning technique to classify the workload by determining a workload-specific fingerprint for the workload. The method includes identifying an optimization metric to optimize during running of the workload in the HPC system. The method includes determining an optimal setting for a plurality of tunable hardware execution parameters as measured against the optimization metric by varying at least a portion of the plurality of tunable hardware execution parameters.

Abstract: Systems and methods described herein make previously stranded power capacity (power that is provisioned for a data center according to a computing system's nameplate power consumption but is currently not useable) available to the data center. Systems described herein generate empirical power profiles that specify expected upper bounds for the power consumption levels that applications trigger. Using the upper bounds for application power-consumption levels, a computing system described herein can reliably release part of its provisioned nameplate power for other systems or data center consumers, reducing the amount of stranded power in a data center. The method described herein avoids performance penalties for most jobs by using sensor measurements made at a rapid rate explained herein to ensure that a system power cap based on running application's measured peak power consumption is reliable with reference to the power capacitance inherent in the computing system.

Abstract: One embodiment provides a system and method for predicting network power usage associated with workloads. During operation, the system configures a simulator to simulate operations of a plurality of network components, which comprises embedding one or more event counters in each simulated network component. A respective event counter is configured to count a number of network-power-related events. The system collects, based on values of the event counters, network-power-related performance data associated with one or more sample workloads applied to the simulator; and trains a machine-learning model with the collected network-power-related performance data and characteristics of the sample workloads as training data 1, thereby facilitating prediction of network-power-related performance associated with a to-be-evaluated workload.

Abstract: Systems and methods are provided for improving statistical and machine learning drift detection models that monitor computing health of a data center environment. For example, the system can receive streams of sensor data from a plurality of sensors in a data center; clean the streams of sensor data; generate, using a machine learning (ML) model, an anomaly score and a dynamic threshold value based on the cleaned streams of sensor data; determine, using the ML model and based on the anomaly score and the dynamic threshold value, a correctness indicator for a first sensor in the plurality of sensors; and using the correctness indicator, correct the first sensor.

Abstract: One embodiment provides a system and method for predicting network power usage associated with workloads. During operation, the system configures a simulator to simulate operations of a plurality of network components, which comprises embedding one or more event counters in each simulated network component. A respective event counter is configured to count a number of network-power-related events. The system collects, based on values of the event counters, network-power-related performance data associated with one or more sample workloads applied to the simulator; and trains a machine-learning model with the collected network-power-related performance data and characteristics of the sample workloads as training data 1, thereby facilitating prediction of network-power-related performance associated with a to-be-evaluated workload.