Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Buffer Management Policy for a Mixed Transaction and Query Environment

IP.com Disclosure Number: IPCOM000105796D
Original Publication Date: 1993-Sep-01
Included in the Prior Art Database: 2005-Mar-20
Document File: 2 page(s) / 100K

Publishing Venue

IBM

Related People

Chung, JY: AUTHOR [+4]

Abstract

Proposed is a buffer management policy for a buffer pool shared by random (e.g., transactions) and sequential (e.g., queries) access streams. Under this policy, the buffer space used by each sequential access stream is bounded by prespecified minimum and maximum sizes. This prevents monopolization of buffer space by a single query. The minimum and maximum sizes can also be changed dynamically.

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

Buffer Management Policy for a Mixed Transaction and Query Environment

      Proposed is a buffer management policy for a buffer pool shared
by random (e.g., transactions)  and sequential (e.g., queries) access
streams.  Under this policy, the buffer space used by each sequential
access stream is bounded by prespecified minimum and maximum sizes.
This prevents monopolization of buffer space by a single query.  The
minimum and maximum sizes can also be changed dynamically.

      In a mixed transaction and query environment, pages brought in
by sequential access streams needs to be bounded to provide
protection for the randomly accessed hot pages.  The randomly
accessed pages should still be managed by LRU replacement policy.  In
[2,3,5]  a buffer management policy is proposed where in addition to
the normal LRU chain some of the buffer pages (the pages brought in
by all sequential streams) are linked through a secondary LRU chain
called "sequential LRU (SLRU) buffer chain".  The total number of
pages in the SLRU chain are limited by some prespecified amount.
This prevents monopolization of buffer space by the sequential access
streams.  However, it provides no mechanism for bounding the buffer
space used by different queries (access streams) or bounding the
buffer space for different relations accessed by sequential streams.
In [1,4]  various query hot set models are proposed that show 1)
buffer hit probability is a discrete function of buffer size, 2)
buffer hit probability is insignificant or zero if the allocated
buffer space is smaller than the query hot set size, and 3) the
increase in buffer hit probability for buffer size beyond the query
hot set size is very small.  Therefore, a buffer management policy
for such environment should provide the capability of bounding (by
minimum and maximum) the allocated buffer space per query per
relation or a group of queries or relations.

      The proposed buffer management policy in addition to
maintaining the normal LRU chain for all pages and a combined
sequential LRU (SLRU) chain for all pages accessed through sequential
access streams, maintains a Private LRU (PLRU) chain for each
sequential access stream that needs to be bounded by a minimum and a
maximum buffer sizes.  Buffers released from a random process are
always placed at the tail of the normal LRU chain.  However, the
buffers released from a sequential process are placed (linked) at the
tails of normal LRU chain, SLRU chain and its PLRU chain.  Running
counts of the number of pages for each of the PLRU chains and for the
SLRU chain are maintained in a table.  Note that if the newly
released...