1971IBM Systems JournalRequires access

Program restructuring for virtual memory

D. J. Hatfield, J. Gerald

Open publisher page 167 citations

Abstract

Our general experience so far has shown the display of a virtual memory use pattern to be a good diagnostic tool. The automatic sector reordering technique brings noticeable improvements in paging performance where there is room for improvement, reducing the necessary working space (for a given number of exceptions) by as much as one-third to one-half. We have worked entirely in the environment of a simple hierarchy of main memory and a uniform speed replacement memory, with all replacement blocks having the same size. We have found that the reordering process, assuming a page size of 4K bytes, also produces improvements on pages of 8K and 2K bytes with the improvements favoring the doubled page size over the halved. This is consistent with a tendency we have noticed in the programs we have examined for better packed memory to favor larger page sizes. But the page size for clustering bears a direct relationship to the program sectors themselves. It has not proved effective to cluster at the physical page size when the average sector size is half as large or more. The optimal page size for a program depends on (besides physical I/O timings) complicated patterns in the use of virtual memory, about which very little is known.

About this research paper

What this paper is about

Our general experience so far has shown the display of a virtual memory use pattern to be a good diagnostic tool. The automatic sector reordering technique brings noticeable improvements in paging performance where there is room for improvement, reducing the necessary working space (for a given number of exceptions) by as much as one-third to one-half. We have worked entirely in the environment of a simple hierarchy of main memory and a uniform speed replacement memory, with all replacement blocks having the same size. We have found that the reordering process, assuming a page size of 4K bytes, also produces improvements on pages of 8K and 2K bytes with the improvements favoring the doubled page size over the halved. This is consistent with a tendency we have noticed in the programs we have examined for better packed memory to favor larger page sizes. But the page size for clustering bears a direct relationship to the program sectors themselves. It has not proved effective to cluster at the physical page size when the average sector size is half as large or more. The optimal page size for a program depends on (besides physical I/O timings) complicated patterns in the use of virtual memory, about which very little is known.

Why it matters

OpenAlex reports 167 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

Our general experience so far has shown the display of a virtual memory use pattern to be a good diagnostic tool. The automatic sector reordering technique brings noticeable improvements in paging performance where there is room for improvement, reducing the necessary working space (for a given number of exceptions) by as much as one-third to one-half. We have worked entirely in the environment of a simple hierarchy of main memory and a uniform speed replacement memory, with all replacement blocks having the same size. We have found that the reordering process, assuming a page size of 4K bytes, also produces improvements on pages of 8K and 2K bytes with the improvements favoring the doubled page size over the halved. This is consistent with a tendency we have noticed in the programs we have examined for better packed memory to favor larger page sizes. But the page size for clustering bears a direct relationship to the program sectors themselves. It has not proved effective to cluster at the physical page size when the average sector size is half as large or more. The optimal page size for a program depends on (besides physical I/O timings) complicated patterns in the use of virtual memory, about which very little is known.

Key concepts: Demand paging, Page, Virtual memory, Page fault, Paging, Computer science, Byte, Memory map

Related papers

Back to paper searchBrowse research topicsOriginal source
Program restructuring for virtual memory — Research Paper | ScholarLens