Browse Prior Art Database

Caching Scheme for Managing Task Migration Sharing

IP.com Disclosure Number: IPCOM000121027D
Original Publication Date: 1991-Jul-01
Included in the Prior Art Database: 2005-Apr-03
Document File: 2 page(s) / 73K

Publishing Venue

IBM

Related People

So, K: AUTHOR [+2]

Abstract

Task migration is crucial to the performance of multiprocessor (MP) systems. Unfortunately, they create a problem for cache-based systems, namely the false sharing problem: when a task is migrated from one processor to another, it introduces an overhead in maintaining cache consistency even for data elements that are private to a single task.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 54% of the total text.

Caching Scheme for Managing Task Migration Sharing

      Task migration is crucial to the performance of
multiprocessor (MP) systems.  Unfortunately, they create a problem
for cache-based systems, namely the false sharing problem: when a
task is migrated from one processor to another, it introduces an
overhead in maintaining cache consistency even for data elements that
are private to a single task.

      Disclosed is a caching scheme that effectively maintains
coherency of data sharing in task-migratable multiprocessors.  This
scheme differentiates real data sharing (multiple processes executing
on different processors) from task-migrated false sharing.  These two
types of sharing are managed separately to achieve the best
performance.  In this article, the following caching scheme is
proposed that effectively handles sharing in cache-based
task-migratable multiprocessors.  In addition to conventional line
tag and state bits in a cache directory, each cache line is attached
a task id (TID) of the task which used the line the last time.  A
cache line can be in the following states: private, shared, modified,
and invalid.  Although it is assumed below that there is a bus
connecting the processors and the caches, this scheme can also be
applied to non-bus interconnections, such as system control units.
This cache coherence scheme works as follows:
1.   Read hit: Give data to the processor.
2.   Read miss: Send TID of the executing task along with the addre...