Browse Prior Art Database

Automatic Priority Group Method Based on Mean Time To Wait

IP.com Disclosure Number: IPCOM000079969D
Original Publication Date: 1973-Oct-01
Included in the Prior Art Database: 2005-Feb-26
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Page, JB: AUTHOR [+2]

Abstract

There has been demonstratable proof over the last 3-4 years that, under most conditions, IBM's Operating Systems perform better (more throughput) if users that perform more input/output are dispatched first and more often than users that "hog" the CPU resource without doing input/output. This creates a better balance between the CPU resource and the input/ output resource. In order to accomplish this, some method must be used to determine the relative degree of "I/O boundedness" or "CPU boundedness" of each user in the system, so they can be ranked in a dispatchable order from most "I/O bound" to most "CPU bound". This description gives one method of accomplishing this ordering.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 78% of the total text.

Page 1 of 1

Automatic Priority Group Method Based on Mean Time To Wait

There has been demonstratable proof over the last 3-4 years that, under most conditions, IBM's Operating Systems perform better (more throughput) if users that perform more input/output are dispatched first and more often than users that "hog" the CPU resource without doing input/output. This creates a better balance between the CPU resource and the input/ output resource. In order to accomplish this, some method must be used to determine the relative degree of "I/O boundedness" or "CPU boundedness" of each user in the system, so they can be ranked in a dispatchable order from most "I/O bound" to most "CPU bound". This description gives one method of accomplishing this ordering.

The method establishes several subgroups. Each subgroup is a distinct range of mean time to wait values and corresponds to a priority range within the ordered dispatchable queue. The method knows how much CPU time each user has used during the last interval, and how many times the user was not ready to use the CPU (this is presumed to be because the user was waiting for I/O). Dividing the CPU time used by the number of waits for I/O gives the mean time to wait value. Example:

Suppose during the last measured time interval, user A used 500 units of time and waited 10 times. Then the mean time to wait is 500/10 or 50. If the user is currently in a subgroup with a range of mean time to wait of 100-150 units, then the user is displayi...