Overview / Usage

General-purpose workloads exhibit patterns throughout their execution on a CPU. The patterns repeat over time due to the presence of loops in the programs. Many have tried to predict patterns for various purposes, namely hardware reconfiguration or simulation points detection. Most of them do it by detecting execution phases from instrumented pieces of source code. One of the questions that I am trying to answer is how to anticipate phase changes when the source information is not available and just looking at the traces of performance metrics.

Methodology / Approach

I used Intel's emon tool to collect performance data from workloads running on Intel's general-purpose CPUs. The traces are part of the data set and transformed into a supervised learning problem.

​

Technologies Used

Specifically, I am using Python and libraries such as numpy, pandas, sklearn, tensorflow and keras to train machine learning algorithms and analyze the data.

I use Intel's DevCloud to train these models.

Repository

https://github.com/esalcort/ML_CPI_predictor

Other links

• Phase-Aware CPU Workload Forecasting
• Learning-based Phase Classification and Prediction with PMC