Browse Prior Art Database

Dynamic Cache Purge

IP.com Disclosure Number: IPCOM000114814D
Original Publication Date: 1995-Feb-01
Included in the Prior Art Database: 2005-Mar-29
Document File: 2 page(s) / 43K

Publishing Venue

IBM

Related People

Christensen, NT: AUTHOR

Abstract

A method for purging the contents of a cache storage without the need for quiescing users of the cache for the duration of the operation.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 70% of the total text.

Dynamic Cache Purge

      A method for purging the contents of a cache storage without
the need for quiescing users of the cache for the duration of the
operation.

      This invention allows purging of a cache storage to proceed
without the need to quiesce the system.  It takes advantage of the
fact that a cache is structure as an array with multiple rows and
columns (congruence classes and associative sets, respectively.)
There are a small number of associative sets, say 4 or 8, and perhaps
thousands of congruence classes.  The invention uses a small
register, called a 'set delete register', This register contains one
bit per associative set.  When a bit is active, accesses to the
corresponding set are restricted.  Specifically, All accesses to
valid entries in a deleted set are processed normally.  However, any
invalid compartments in a 'deleted' set are removed from
consideration for reassignment on future cache misses.

      The purge process proceeds through the associative sets
sequentially, and within each set processes each congruence class
sequentially as well.  As each congruence class is purged the normal
castout mechanisms ensure that all stores to the Line are completed,
that no CPs own the data being cast out, and that the compartment is
finally invalidated.  It can be seen that as the purge progresses all
the compartments in a set become invalidated, until eventually the
entire set has become purged, and truly 'deleted'.  The system
con...