Performance Tuning Using Standardized Performance Tests and Machine Learning

Abstract

Computer systems have a large number of hardware and operating system configuration parameters that can be critical to the performance of a service or application running on it. Experts manually tune these parameters based on their intuition and experience, which can be very time-consuming and inefficient. While there is a lot of work done in modifying configurations of an application to predict and improve its performance, these strategies are based upon tuning the configuration parameters pertaining to that application. Red Hat has compiled a large repository of results of running the Red Hat open-source performance measurement tool pbench on various configurations of the Red Hat linux operating system, with the hope of basing an expert system for performance tuning upon this data. We plan to use this repository to tune the performance of the underlying operating system and thus improve application performance at the same time. We plan to use unsupervised machine learning to categorize the different configurations and workloads present in pbench data, and use the set of performance results to recommend the optimal configuration for a new system based on its workload of interest and application type, as part of an expert system for linux performance tuning. Thus, rather than tuning one specific application, we seek to generally tune the operating system to improve performance for a wide range of applications that might run on that operating system.

Join meeting in Sococo VH 209. Login: tuftscs.sococo.com

Join Zoom Meeting: https://tufts.zoom.us/j/98610939077

PASSWORD: See colloquia email

Dial by your location: +1 646 558 8656 US (New York)

Meeting ID: 986 1093 9077

Passcode: See colloquia email