Browse Prior Art Database

Evaluating Advanced Time Sharing Systems

IP.com Disclosure Number: IPCOM000075574D
Original Publication Date: 1971-Oct-01
Included in the Prior Art Database: 2005-Feb-24
Document File: 4 page(s) / 37K

Publishing Venue

IBM

Related People

Florkowski, JH: AUTHOR

Abstract

A general purpose analytic model determines resource utilizations, mean queue lengths, response times, throughput rates, etc. for closed systems having a finite quantity of requestors competing for the system's resources. Input parameters representing the hardware configuration, distribution of I/O events, service times of the resources, and dynamic system control variables are mapped into the model which accurately tracks the measured system.

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 38% of the total text.

Page 1 of 4

Evaluating Advanced Time Sharing Systems

A general purpose analytic model determines resource utilizations, mean queue lengths, response times, throughput rates, etc. for closed systems having a finite quantity of requestors competing for the system's resources. Input parameters representing the hardware configuration, distribution of I/O events, service times of the resources, and dynamic system control variables are mapped into the model which accurately tracks the measured system.

Resource types 1 through 4 of a simplified, interacting model structure of a complex time-shared computer are the CPU, paging drums, auxiliary paging disks, and external paging disks, respectively. The model requires three statistics per resource type, i: (1) n(i): number of resources, (2) F(i): fraction of traffic intensity, lambda (lambda), and (3) T(i): service time of the resource. Additionally, it requires the total request population (multiprogramming tasks), R.

n(i),the number of resources for each resource type, is simply a configuration variable. For example, the number of auxiliary disk drives may be four, so that n(3)=4. Only one resource type is used to represent these four disk drives, with the assumption that all of the traffic intensity to the 3rd resource type is equally distributed over the n(3) resources.

F(i), the fraction of traffic intensity for the ith resource type, is obtained for a paging system (4096 bytes/page) as follows: The paging operations per I/O device are tabulated and then the fractional distribution, F(i), of paging activity is determined by dividing each devices activity by the total activity. For nonpaging systems, the quantity of I/O accesses would produce these fractions along with a mean block size of transferred bytes per access. Whether the system is paged or not, the CPU fraction, F(1), is unity since all I/O events return to this one common resource.

T(i), the service time, is calculated for the I/O devices for both paging and nonpaging systems. This is obtained from measurements of the actual access time distribution or the use of an assumed distribution (e.g., uniform) and knowledge of the device's transfer characteristics.

The computer service time, T(1), is derived through equation (1). The required CPU time between I/O requests for a task is a composite of task (problem program) time and supervisor overhead time. T(1)=PP + F(2)(oh(1)) + (F(3)+ F(4)) (oh(2)) + (oh(3)+ oh(4))/TP (1) where: TP: problem program I/O page requests during a time slice PP: problem program time per page oh(1) : supervisor overhead per drum page oh(2) : supervisor overhead per disk page oh(3) : supervisor overhead per time slice start oh(4) : supervisor overhead per time slice end F(2), F(3), F(4) : traffic intensity fractions.

The problem program time, PP, is the CPU time between I/O paging events and is obtained from statistics gathered by the system, or from a trace of the problem program. The drum overhead time (oh(1)) to p...