Browse Prior Art Database

Multilevel Staging Memory Control to Prevent Data Loss

IP.com Disclosure Number: IPCOM000088804D
Original Publication Date: 1977-Aug-01
Included in the Prior Art Database: 2005-Mar-04
Document File: 3 page(s) / 17K

Publishing Venue

IBM

Related People

Mayka, RJ: AUTHOR

Abstract

This article describes a method of precluding data loss during movements of data from one level of a hierarchic store or memory to another. This method consists of the explicit sequencing of certain events during the data movement process to eliminate data loss as a potential result of system failures.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 34% of the total text.

Page 1 of 3

Multilevel Staging Memory Control to Prevent Data Loss

This article describes a method of precluding data loss during movements of data from one level of a hierarchic store or memory to another. This method consists of the explicit sequencing of certain events during the data movement process to eliminate data loss as a potential result of system failures.

A hierarchic store is a series of "n" memory levels arranged in priority order. The priority of a particular memory level is governed by its rate of access: the faster the access, the higher the priority. Typically, faster access media contain less capacity than slower access media. This is because access time is inversely proportional to cost. Access to a hierarchic store always begins with the highest priority memory level, Thus, the search for a particular data item always begins on the fastest access medium. If the data item is not found, the second highest priority memory level is searched, and so on.

For the hierarchic store to operate efficiently, the data items currently being accessed should reside in the highest priority memory level and, thus, on the fastest access media. Therefore, whenever a data item is referenced, it is brought from its current level in the store to the highest level (fastest access media). This upward migration is usually done on a level-by-level basis. When a data item falls into disuse, it is no longer efficient to retain it in the highest level of the store. The data item is then migrated to lower and lower levels of the store based on some type of aging criteria.

The use of such a hierarchic store results in a great deal of data movement between levels of the store. Usually, a directory of some sort is associated with each level of the store. Assume that all required information, such as directories and space maps, is located on nonvolatile media. Thus, interlevel data movement also involves directory updates, space allocations and deallocations, and space map updates.

A system failure in a hierarchic store is any failure which causes immediate cessation of activity within the store. A CPU failure or a power failure are examples of a system failure. If a system failure should occur while data is being moved between levels, the data loss that might result could be disastrous. Data loss must be minimized.

The activities associated with data movement cannot be performed atomically. Physical limitations require that these activities be performed in some order.

The following is an overview scenario depicting an activity sequence which precludes data loss in the event of a system failure. 1. Allocate space on target media. 2. Search source directory and read source data. 3. Write data on target media. 4. Insert entry into target directory. 5. Delete entry from source directory.
6. Deallocate space on source media.

Steps 1 through 6 are ordered such that a system failure occurring anywhere within this sequence will not result in data loss. In all ca...