Computing resources workload scheduling

Patent metadata

Field	Value
Publication number	US-9959146-B2
Application number	US-201615089481-A
Country	US
Kind code	B2
Filing date	Apr 2, 2016
Priority date	Oct 20, 2015
Publication date	May 1, 2018
Grant date	May 1, 2018

Abstract

Examples may include techniques to a schedule a workload to one or more computing resources of a data center. A class is determined for the workload based on a workload type or profile for the workload. Predicted operating values for at least one of the one or more computing resources is determined based on the class and the predicted operating values are used as inputs in at least one scoring model to evaluate the workload being supported by the at least one of the one or more computing resources. The workload is then scheduled to the at least one or more computing resources based on the evaluation.

First claim

The invention claimed is:

1. An apparatus comprising:
circuitry communicatively coupled to a data center;
a distributor for execution by the circuitry to receive an indication of a workload distribution for a workload to be supported by one or more computing resources of a data center;
a classify component for execution by the circuitry to determine a class for the workload based on a workload type or profile for the workload;
a modeler for execution by the circuitry to determine predicted operating values for at least one of the one or more computing resources based on the class and input the one or more predicted operating values in at least one scoring model to evaluate the workload being supported by the at least one of the one or more computing resources;
a scheduler for execution by the circuitry to schedule the workload to the at least one of the one or more computing resources based on the evaluation;
a collector for execution by the circuitry to gather training operating information for one or more workloads included in the workload type or profile while the one or more computing resources of the data center support the one or more workloads, the training operating information including an inlet or an outlet temperature for a platform or rack housing the one or more computing resources, a power consumption for a platform or rack housing the one or more computing resources, processor cache miss information, network data throughput latency information, memory access latency information, throttling activation information for the one or more computing resources, margin to a peak operating temperature threshold for the one or more computing resources, or a volumetric airflow for a platform and/or rack housing the one or more computing resources; and
a store for execution by the circuitry to store the gathered training operating information, the classify component to use the stored and gathered training operating information to:
cluster group the training operating information via k-means clustering;
learn a classifier based on the cluster group of the training operating information via a support vector machine (SVM); and
assign a regression fit function to classified training operating information classified by the classifier based on a least squares approach.

2. The apparatus of claim 1 , the modeler to determine the one or more predicted operating values based on the class comprises the modeler to use the regression fit function assigned to the classified training operating information to determine the one or more predicted operating values.

3. The apparatus of claim 1 , the workload type or profile comprising one of a first workload type or profile that is processing or processor intensive, a second workload type or profile that is memory intensive, a third workload type or profile that is network switch intensive, a fourth workload type or profile that is storage intensive or a fifth workload type or profile that is a balanced workload type or profile that has relatively equal processor, memory, network switch and storage intensities.

4. The apparatus of claim 1 , comprising a digital display coupled to the circuitry to present a user interface view.

5. A method comprising:
receiving, at a processor circuit, an indication of a workload distribution for a workload to be supported by one or more computing resources of a data center;
determining a class for the workload based on a workload type or profile for the workload;
determining predicted operating values for at least one of the one or more computing resources based on the class and inputting the one or more predicted operating values in at least one scoring model to evaluate the workload being supported by the at least one of the one or more computing resources;
scheduling the workload to the at least one of the one or more computing resources based on the evaluation;
gather training operating information for one or more workloads included in the workload type or profile while the one or more computing resources of the data center support the one or more workloads, the training operating information including an inlet or an outlet temperature for a platform or rack housing the one or more computing resources, a power consumption for a platform or rack housing the one or more computing resources, processor cache miss information, network data throughput latency information, memory access latency information, throttling activation information for the one or more computing resources, margin to a peak operating temperature threshold for the one or more computing resources, or a volumetric airflow for a platform and/or rack housing the one or more computing resources;
cluster grouping the training operating information using k-means clustering;
learning a classifier based on the cluster grouping of the training operating information using a support vector machine (SVM);
assigning a regression fit function to classified training operating information classified by the classifier based on a least squares approach; and
using the regression fit function assigned to the classified training operating information to determine the one or more predicted operating values.

6. The method of claim 5 , the workload type or profile comprising one of a first workload type or profile that is processing or processor intensive, a second workload type or profile that is memory intensive, a third workload type or profile that is network switch intensive, a fourth workload type or profile that is storage intensive or a fifth workload type or profile that is a balanced workload type or profile that has relatively equal processor, memory, network switch and storage intensities.

7. The method of claim 5 , determining predicted operating values based on the class comprises using the regression fit function assigned to the classified training operating information to determine the one or more predicted operating values.

8. At least one non-transitory machine readable medium comprising a plurality of instructions that in response to being executed by a system causes the system to:
receive an indication of a workload distribution for a workload to be supported by one or more computing resources of a data center;
determine a class for the workload based on a workload type or profile for the workload;
determine predicted operating values for at least one of the one or more computing resources based on the class and input the one or more predicted operating values in at least one scoring model to evaluate the workload being supported by the at least one of the one or more computing resources;
schedule the workload to the at least one of the one or more computing resources based on the evaluation;
gather training operating information for one or more workloads included in the workload type or profile while the one or more computing resources of the data center support the one or more workloads, the training operating information including an inlet or an outlet temperature for a platform or rack housing the one or more computing resources, a power consumption for a platform or rack housing the one or more computing resources, processor cache miss information, network data throughput latency information, memory access latency information, throttling activation information for the one or more computing resources, margin to a peak operating temperature threshold for the one or more computing resources, or a volumetric airflow for a platform and/or rack housing the one or more computing resources;
cluster grouping the training operating information using k-means clustering;
learn a classifier based on the cluster grouping of the training operating information using a support vector machine (SVM);
assign a regression f

Assignees

• Intel Corp

Inventors

• Ahuja Nishi
• Khanna Rahul
• Daniel Abishai
• Fu Diyong

Classifications

• G06F9/5094

  where the allocation takes into account power or heat criteria (power management in computers in general G06F1/3203; thermal management in computers in general G06F1/206) · CPC title

• G06F2209/501

  Performance criteria · CPC title

• G06N99/005

  Physics · mapped topic

• G06F9/505 Primary

  considering the load · CPC title

• Y02B60/142

  Cross-Sectional Technologies · mapped topic