Browse Prior Art Database

Biasing Cache Threshold Pointers Toward Less Pre-Staging

IP.com Disclosure Number: IPCOM000034301D
Original Publication Date: 1989-Jan-01
Included in the Prior Art Database: 2005-Jan-27
Document File: 5 page(s) / 20K

Publishing Venue

IBM

Related People

Oliver, JK: AUTHOR

Abstract

A technique is described whereby computer cache storage performance is improved by biasing cache threshold pointers toward less pre-saging. The concept adjusts the threshold pointers so that when the hit ratio varies little, or not at all, over the entire range of threshold pointers, a smaller, average effective block length is used. This results in less pre-staging and, therefore, increases the performance of cache operations. In response to the need for greater performance many disk attachments implement cache algorithms to try to maintain the most recently or most often requested data directly in the cache memory. When the operating system requests data, many times the request can be met entirely with data contained in the cache, thereby reducing access time.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 24% of the total text.

Page 1 of 5

Biasing Cache Threshold Pointers Toward Less Pre-Staging

A technique is described whereby computer cache storage performance is improved by biasing cache threshold pointers toward less pre-saging. The concept adjusts the threshold pointers so that when the hit ratio varies little, or not at all, over the entire range of threshold pointers, a smaller, average effective block length is used. This results in less pre-staging and, therefore, increases the performance of cache operations. In response to the need for greater performance many disk attachments implement cache algorithms to try to maintain the most recently or most often requested data directly in the cache memory. When the operating system requests data, many times the request can be met entirely with data contained in the cache, thereby reducing access time. Techniques have been implemented in the past which improve the alignment of cache blocks with a short sequence pattern of disk references, called pre-staging. Here, one or more extra blocks of data are entered into cache when a read miss occurs. This pre-staging is performed in the hopes that the user is reading sequentially and will request data from those blocks in the near future. Essentially, the pre-staging of the data is in anticipation of the possibility of saving time in the future. The read requests would then come directly from cache, resulting in a gain in performance. However, occasionally the user is not reading sequentially and, therefore, does not request data from the pre-staged blocks. In this case, the extra time required to read the pre-staged data from the disk is a detriment to the performance. This prior technique also explained how different values of cache threshold pointers resulted in different average effective block lengths. Another technique concentrated on the automatic adjustment of disk cache operation based on tracking the read hit ratios. Here, a means was provided whereby the cache threshold pointers were adjusted based on the read hit ratio. The threshold pointers were periodically adjusted and the cache hit ratio was measured at each value of the threshold pointers. In this manner, the threshold pointers were adjusted to keep the hit ratio at a maximum. However, a problem existed in this concept in situations where the hit ratio did not vary significantly over the entire range of values of threshold pointers. When the average effective block length was increased by varying the threshold pointers, the time required to read the pre-staged data from the disk increased. If the hit ratio does not rise sufficiently so as to compensate for this extra time, the operational performance decreased. The concept described herein attempts to overcome the various deficiencies by concentrating on the optimization of cache performance. This is done by biasing the threshold pointers toward less pre-staging.

Algorithms are used to take into account the performance gain or loss, caused by the change...