mistat (version 2.0.4)

COMPURESP: Computer Response Time Optimization

Description

The experiment described here was part of an extensive effort to optimize a UNIX operating system.

Usage

data(COMPURESP)

Format

A data frame with 18 observations on the following 10 variables.

F

a factor with levels 1 2, representing KMCs used

B

a factor with levels 1 2 3, representing File Distribution

C

a factor with levels 1 2 3, representing Memory Size

D

a factor with levels 1 2 3, representing System Buffers

E

a factor with levels 1 2 3, representing Sticky Bits

A

a factor with levels 1 2 3, representing Disk Drives

G

a factor with levels 1 2 3, representing INODE Table Entries

H

a factor with levels 1 2 3, representing Other System Tables

Mean

mean time taken for the system to complete commands execution

SN

S/N ratio \(\eta = -10 log_{10}(\frac{1}{n} \sum_i y_i^2 )\)

Details

The experiment described here was part of an extensive effort to optimize a UNIX operating system running on a VAX 11-780 machine. The machine had 48 user terminal ports, two remote job entry links, four megabytes of memory, and five disk drives. The typical number of users logged on at a given time was between 20 to 30.

1. Problem Definition. Users complained that the system performance was very poor, especially in the afternoon. The objective of the improvement effort was to both minimize response time and reduce variability in response.

2. Response variable. In order to get an objective measurement of the response time two specific representative commands called ‘standard’ and ‘trivial’ were used. The ‘standard’ command consisted of creating, editing and removing a file. The ‘trivial’ command was the UNIX system ‘date’ command. Response times were measured by submitting these commands every 10 minutes and clocking the time taken for the system to complete their execution.

Examples

data(COMPURESP)

layout(matrix(1:4, 2, byrow=TRUE))

with(COMPURESP,
     interaction.plot(
       x.factor=F, 
       trace.factor=rep(0, length(F)), 
       response=SN, 
       legend=FALSE,
       type="b",
       pch=15:18,
       ylim=c(-17, -10)))

with(COMPURESP,
     interaction.plot(
       x.factor=B, 
       trace.factor=rep(0, length(B)), 
       response=SN, 
       legend=FALSE,
       type="b",
       pch=15:18,
       ylim=c(-17, -10)))

with(COMPURESP,
     interaction.plot(
       x.factor=C, 
       trace.factor=rep(0, length(C)), 
       response=SN, 
       legend=FALSE,
       type="b",
       pch=15:18,
       ylim=c(-17, -10)))

with(COMPURESP,
     interaction.plot(
       x.factor=D, 
       trace.factor=rep(0, length(D)), 
       response=SN, 
       legend=FALSE,
       type="b",
       pch=15:18,
       ylim=c(-17, -10)))

layout(1)