Browse Prior Art Database

Method of estimating the number of transaction and CPU utilization of peak day using index number for peak day and multiple linear regression analysis.

IP.com Disclosure Number: IPCOM000018700D
Original Publication Date: 2003-Aug-01
Included in the Prior Art Database: 2003-Aug-01
Document File: 4 page(s) / 26K

Publishing Venue

IBM

Abstract

It is necessary to predict the future of a capacity in the system which a large-scale calculation system transaction like a financial institution generates for the stability of a system. Prediction of the transaction generating number of cases was estimated by a past actual result value and a past annual rate. Moreover, the multiplication of the value of the path length calculated for every subsystem was carried out to future number-of-cases prediction, and prediction of the rate of CPU use estimated it to be it. Subjects are the accuracy of the prediction transaction number of cases, and the accuracy of path length. As for the actual number of cases of a peak day, the element "monthly" and "according to day of the week" is added. Therefore, in consideration of an "annual rate", prediction was difficult. Moreover, the rate of CPU use and path length were difficult to estimate it for every subsystem estimating in many cases from the test result of the MIPS value and the rate of CPU use of a machine etc. in the acting-before-the-audience system which has already worked. The point using the statistics technique of computing "the characteristic index of a prediction day" and the constant of a path length element by "multiplex regression analysis" to a capacity management and the prediction technique has this invention. a result -- a future prediction value -- the system improved sharply Moreover, the input of these tools of analysis is only the past actual result value. Therefore, it contributes also to curtailment of the work load spent on analysis compared with the technique searched for from the conventional test intention.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 33% of the total text.

Page 1 of 4

  Method of estimating the number of transaction and CPU utilization of peak day using index number for peak day and multiple linear regression analysis.

0. Background It is necessary to predict the future of a capacity in the system which a large-scale calculation system transaction like a financial institution generates for the stability of a system. Prediction of the transaction generating number of cases was estimated by a past actual result value and a past annual rate. Moreover, the multiplication of the value of the path length calculated for every subsystem was carried out to future number-of-cases prediction, and prediction of the rate of CPU use estimated it to be it. Subjects are the accuracy of the prediction transaction number of cases, and the accuracy of path length. As for the actual number of cases of a peak day, the element "monthly" and "according to day of the week" is added. Therefore, in consideration of an "annual rate", prediction was difficult. Moreover, the rate of CPU use and path length were difficult to estimate it for every subsystem estimating in many cases from the test result of the MIPS value and the rate of CPU use of a machine etc. in the acting-before-the-audience system which has already worked.

The point using the statistics technique of computing "the characteristic index of a prediction day" and the constant of a path length element by "multiplex regression analysis" to a capacity management and the prediction technique has this invention. a result -- a future prediction value -- the system improved sharply Moreover, the input of these tools of analysis is only the past actual result value. Therefore, it contributes also to curtailment of the work load spent on analysis compared with the technique searched for from the conventional test intention.

1. Advancement of Prediction Number of Cases

A peak day is defined beforehand (for example, the 25th day, the end of the month). The characteristic by monthly [ of a peak day ] is evaluated as a "monthly index", and the characteristic by the day of the week of a peak day is evaluated as "an index according to day of the week." Evaluation of each index considers the actual result value of the peak day number of cases classified by subsystem as an input, and computes it by the following methods. An "annual rate" is computed from the number of cases of the same month from which a year is different, and the same day of the week. After performing annual rate compensation (removal of the element by the annual rate) for the accumulated past actual result data from an "annual rate", the day-of-the-week pattern characteristic is removed, and a "monthly index" is computed. Removal of the day-of-the-week pattern characteristic is rectified in quest of the multiple for comparing by the data which overlap by "monthly" by the data for every day-of-the-week pattern group, and arranging average value. By the "annual rate" and a "monthly index", the data which remov...