Browse Prior Art Database

Quck discard for file pages

IP.com Disclosure Number: IPCOM000012884D
Original Publication Date: 2003-Jun-06
Included in the Prior Art Database: 2003-Jun-06
Document File: 1 page(s) / 36K

Publishing Venue

IBM

Abstract

This article addresses doing a one pass scan, instead of a two pass scan when large amounts of pages for files are being brought into memory, to speed up memory availability.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 100% of the total text.

Page 1 of 1

Quck discard for file pages

We have a problem with persistent pages which are scanned twice by the LRU when a page needs to be freed. In a workload where large number of persistent pages are brought into memory, inevitably we reach a point where we need to free pages in order to make room for more pages.

Currently LRU operates by clearing a reference bit and then checks on the next pass to see it that bit is still clear. If it is clear it removes the page. In an environment where reuse of file pages is desirable we want to use the Virtual Memory Manager file cache, however as it is often necessary to free large numbers of these pages quickly, the cost of doing this two-pass scan on the persistent pages is considerable.

Our proposal is to give applications which do large file I/O the ability to open some/all of their file with a quick discard option. Persistent pages which are brought in from a file opened in this mode would be discardable (freed) on the first pass, not on a second pass. This file option could be an additional open flag or as a mount option on the file system.

Any persistent page which is associated to a file opened with quick discard is available to be freed on the first pass of LRU. This would significantly increase the effect of the page replacement algorithm while still maintaining the advantages of a large file cache.

Disclosed by International Business Machines Corporation

1