Browse Prior Art Database

Performance Indicator recommender system with performance data using piece-wise linear regression

IP.com Disclosure Number: IPCOM000198333D
Publication Date: 2010-Aug-05
Document File: 2 page(s) / 72K

Publishing Venue

The IP.com Prior Art Database

Abstract

Rajan Ravindran, Ramakrishnan Kannan, Anbazhagan Mani, Karthik Subbian Performance Indicator recommender system with performance data using piece-wise linear regression. We run piece-wise linear regression on the historic data of performance of a system, to identify the relevant modules that affects the performance of the system.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 70% of the total text.

Page 1 of 2

Performance Indicator recommender system with performance data using piece -wise linear regression

For performance tuning in a system, there are various performance indicator available such as module versions, environment setting etc. The two major techniques that are current followed to improve performance in a system is changing module version / environment settings. On each product build, there is an automated performance test script which will be executed by changing these various parameters and identifies the best performance by tuning these performance indicators. Typically in customer environment where there are multiple product, hardware etc exists such recommender system will be potentially big use to the customer.

Problem

In the above scenario, the performance results of these tests are updated in the database. If there is a performance lag or improvement the identification of the performance indicator related to this performance change is a manually done and is a challenge. We are proposing a recommender system using piece-wise linear regression that recommends the possible list of modules/environment changes that might be related to the performance lagging/improvement.

Example

Let the list of modules of the product be a1,…,an. Let on the 1st build the performance is 1000 ms.

Weekly Build

Component Changed from previous build

Performance (in ms)

Performance Change from previous build (in ms)

2nd Week

A3,a4,a5

1250

-250

3rd Week

A4,a6,a7

1450

-200

4th...