Browse Prior Art Database

Blocked Page-Out

IP.com Disclosure Number: IPCOM000041630D
Original Publication Date: 1984-Feb-01
Included in the Prior Art Database: 2005-Feb-02
Document File: 2 page(s) / 14K

Publishing Venue

IBM

Related People

Beretvas, T: AUTHOR [+6]

Abstract

The new concept is to improve the total VM system performance by allocating contiguous slots for not necessarily related (demand) page-outs up to a device-dependent limit, i.e., blocking page-outs. The new concept can result in writing multiple pages per SIO. In a demand paging system control program (SCP), when a page frame shortage is detected (number of page frames free plus number of pages being written out, compared to a threshold), a "table scan" mechanism is invoked to replenish the available frame list. This cyclic scan of the table looks for available page frames, and since the ordering of the core table is by real storage address, it affects users in a random fashion. The core table scan takes unreferenced pages in its path, until the threshold is met.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 1 of 2

Blocked Page-Out

The new concept is to improve the total VM system performance by allocating contiguous slots for not necessarily related (demand) page-outs up to a device- dependent limit, i.e., blocking page-outs. The new concept can result in writing multiple pages per SIO. In a demand paging system control program (SCP), when a page frame shortage is detected (number of page frames free plus number of pages being written out, compared to a threshold), a "table scan" mechanism is invoked to replenish the available frame list. This cyclic scan of the table looks for available page frames, and since the ordering of the core table is by real storage address, it affects users in a random fashion. The core table scan takes unreferenced pages in its path, until the threshold is met. Since the usual shortfall is only one page frame, it is usual to acquire only one page each time. If the page taken was changed (from the previous copy on secondary storage), then it has to be written out onto a pageing DASD (direct-access storage device) before the frame is made available for re-use. Thus, normally one page is written per SIO at a time. SCP algorithms for device and slot allocation tend to maximize the parallelism among identical devices (one page allocated to each device in rotation), the so-called round robin algorithm. This is done to increase the overlap if there are identical devices, but decreases the I/O chaining that can be accomplished. Whenever a new page-out request arrives, a new device is selected for the page-out. I/O chaining occurs if (and only if) the requests arrive faster than the device can process them. Thus, page reads can be chained with page-out requests, since the arrival of page read requests is independent from the round robin algorithm, but write requests are not chained very often. The concept disclosed here is to select multiple (preferably contiguous) slots from the same device, switching to the next available device only after a device-dependent limit has been reached. The device-dependent limit is determined in one of two ways: 1. Each additional request within the limit represents a (small) linear increment in the device access time. Any additional request beyond the limit represents a non-linear increase in the access time. An example of such a limit is the cylinder limit of DASD, where going beyond the cylinder limit increases the access time non-l...