Technologies for predicting computer hardware performance with machine learning
Technologies for predicting computer hardware performance with machine learning are disclosed. Analysis of telemetry data through machine learning and statistical modeling is used to determine whether various components of a compute device such as a fan or memory are failing or are otherwise potentially impacting performance of the compute device. For example, machine-learning-based algorithms may be used to determine an impact of a latency of memory accesses may have on time to execute workloads.
1. An orchestration system for managing failure of a fan of a compute device, the orchestration system comprising:
fan performance classification circuitry to:
access parameters of an algorithm for determination of a likelihood of failure of the fan;
receive telemetry data of the compute device, wherein the telemetry data comprises speeds of a plurality of fans of the compute device; and
analyze, with use of the parameters of the algorithm for determination of a likelihood of failure of the fan, the telemetry data to generate an output of the algorithm for determination of a likelihood of failure of the fan, wherein the output of the algorithm for determination of a likelihood of failure of the fan indicates a likelihood of failure of a fan of the plurality of fans of the compute device, wherein the output of the algorithm is a difference between a statistical value of a fan speed and an F value of an F-distribution; and
workload allocation circuitry to:
determine, and based on the output of the algorithm for determination of a likelihood of failure of the fan, whether action should be taken to mitigate an effect failure of the fan; and
move, by the orchestration system and in response to a determination that action should be taken to mitigate an impact of failure of the fan, a workload of the compute device to a different compute device.
2. The orchestration system of claim 1 , wherein the parameters of the algorithm for determination of a likelihood of failure of the fan comprises parameters of a statistical model,
wherein to analyze, with use of the parameters of the algorithm for determination of a likelihood of failure of the fan, the telemetry data to generate an output of the algorithm for determination of a likelihood of failure of the fan comprises to analyze, by the orchestration system and with use of the parameters of the statistical model, the telemetry data to generate an output of the statistical model.
3. The orchestration system of claim 2 , further comprising:
fan performance classification training circuitry to:
access telemetry data of training workloads; and
calculate, with use of the telemetry data of the training workloads, a mean fan speed vector and a covariance matrix, wherein the parameters for the statistical model comprise the mean fan speed vector and the covariance matrix.
4. The orchestration system of claim 3 , wherein to analyze the telemetry data to generate an output of the algorithm for determination of a likelihood of failure of the fan comprises:
calculate, with use of the statistical model, a T 2 statistical value of the fan speed,
calculate, with use of the statistical model, the F value of the F-distribution; and
determine a difference between the T 2 statistical value and the F value.
5. The orchestration system of claim 1 , wherein the parameters of the algorithm for determination of a likelihood of failure of the fan comprises parameters of a naïve Bayesian classifier, wherein to analyze, with use of the parameters of the algorithm for determination of a likelihood of failure of the fan, the telemetry data to generate an output of the algorithm for determination of a likelihood of failure of the fan comprises to analyze, by the orchestration system and with use of the parameters of the naïve Bayesian classifier, the telemetry data to generate an output of the naïve Bayesian classifier.