Browse Prior Art Database

Method and System for Environment Aware Maintenance Task Optimization based on Machine Learning Disclosure Number: IPCOM000250426D
Publication Date: 2017-Jul-13
Document File: 4 page(s) / 511K

Publishing Venue

The Prior Art Database


This article describes a method for automatic scheduling of maintenance operations in times of low system utilization. The maintenance windows are determined without human interaction by applying machine learning techniques on a history of observed workload statistics.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 37% of the total text.

Method and System for Environment Aware Maintenance Task Optimization based on Machine Learning

Background and Motivation Disclosed is a system that solves the problem of scheduling maintenance and production tasks that potentially interfere with each other and compete for system resources. This situation is quite common for, e.g., database management systems, which run data transactions (insert/update/delete records) as well as reading queries as production workloads, and need maintenance tasks (indexing, building statistics, data reorganization, data compression, etc.) in order to keep the data storage efficient. Whenever such operations interfere with each other, performance of critical production workloads can decrease tremendously, causing issues in various business applications. Automatic Scheduling of Maintenance Operations based on a Workload History The disclosed system monitors system utilization parameters, such as CPU, memory, disk, and network utilization, and will record those utilization statistics. By analysing the historical usage data with machine learning algorithms, potential maintenance windows can be predicted, where it is expected that little to no production workloads are active on the system and, thus, resource-demanding maintenance tasks can be scheduled. The more workload data is available, the better the prediction will work. Thus, the system is continuously monitored and the learned models are updated frequently. By following the method that is suggested in this article, the task to determine appropriate maintenance windows can either be completely automated without additional administrative interactions. Alternatively, the output of the algorithm can be presented as suggestions via wizards for manual maintenance window scheduling by system administrators that do not want to lose control over the operations that are running in their environment. In summary, by implementing the solution: • a better performance for productions tasks can be achieved • a better performance for maintenance tasks can be achieved • a better overall system utilization can be achieved • lower system administration overheads are required

Implementation Details The architecture of the disclosed system is illustrated in the figure. Existing components are shaded in white, new/modified components are shaded in green. The disclosed system comprises a state-of-the-art monitoring component, e.g., database performance monitoring tools or system monitoring tools like profilers, and the enhanced scheduler. The monitoring component constantly receives relevant performance metrics, such as CPU and memory utilization, disk or network I/O rates, etc. These metrics are observed by the enhanced scheduler, e.g., by polling or call- back notification mechanisms, and, together with the current time the measure was taken, is used as input for building a workload prediction model. For latter, arbitrary state-of- the-art algorithms that are well-known from cont...