Browse Prior Art Database

Extended Swap Control for Paged Data

IP.com Disclosure Number: IPCOM000045121D
Original Publication Date: 1983-Feb-01
Included in the Prior Art Database: 2005-Feb-06
Document File: 4 page(s) / 79K

Publishing Venue

IBM

Related People

Beretvas, T: AUTHOR [+3]

Abstract

At the expense of some additional data transfer, system and user performance may be improved by the extended swap data handling shown in Fig. 1. It shows the MVS (multiple virtual storage) page swapping operations between main storage and auxiliary storage (e.g., DASD (direct-access storage device)) organized into five types of pages in a user's address space in a swap data set 14 and a page data set 15, called an "extended swap", since it diverts all (or almost all) of the paging traffic caused by swaps into swap datasets.

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 41% of the total text.

Page 1 of 4

Extended Swap Control for Paged Data

At the expense of some additional data transfer, system and user performance may be improved by the extended swap data handling shown in Fig. 1. It shows the MVS (multiple virtual storage) page swapping operations between main storage and auxiliary storage (e.g., DASD (direct-access storage device)) organized into five types of pages in a user's address space in a swap data set 14 and a page data set 15, called an "extended swap", since it diverts all (or almost all) of the paging traffic caused by swaps into swap datasets.

This extended swap transfers out of main storage to DASD the entire swap-in set, i.e., all the referenced pages whether changed or unchanged, so that at the next swap-in time the entire swap-in set can be transferred to main storage from DASD in a single sequential operation from one swap dataset as an entity, in a single stage operation without the possibility of RPS (rotational position sensing) misses between read operations.

The recently unreferenced pages are still most efficiently handled as a page data set because only a small percentage of its pages are changed, so that only the changed unreferenced pages are swapped out into the page data set.

The extended swap-out process is still a two-stage process: the trimmer changed pages are written to the page dataset, and then the other pages are written into the swap dataset. However, the swap-in process is a one-stage process in which only the swap dataset is swapped in the LSQA (local system queue area). Page slots are not used anymore to record the location of the nonLSQA pages written to the swap dataset accesses to the page data set which are made by page faults.

The effect of the extended swap method is to move most of the swap traffic to swap datasets. Paging traffic for page datasets is reduced. Generally, only short page dataset channel programs are used.

Address space 10 in Fig. 2 illustrates five types of pages which ay exist in any MVS user's address space, which are subject to both the swap-out and swap-in processes. Previously in MVS, the LSQA pages were swapped out to a swap data set 11 on DASD, and the other four types of pages were contained in a page data set 12 on DASD. Only changed pages in the address space are swapped out, because unchanged pages already exist in the page data set on DASD. At swap-out time, all changed pages are scheduled to be written to the page dataset (stage one of the swapout operation), followed by the writing of the LSQA pages to the swap dataset (stage two) into one or more empty consecutive slots. LSQA pages contain the auxiliary storage location of pages in the address space. At swap-in time, the swap set for the address space must be read into main storage. In stage one, the LSQA pages are read from the swap dataset. In stage two, the four other types of pages in the working set are read into main storage.

A system control program, such as the IBM MVS (multiple virtual stor...