Browse Prior Art Database

Page Locking Assist for Multisystems Data Sharing Disclosure Number: IPCOM000122199D
Original Publication Date: 1991-Nov-01
Included in the Prior Art Database: 2005-Apr-04
Document File: 3 page(s) / 137K

Publishing Venue


Related People



Disclosed is a technique for providing hardware lookaside support for improving the efficiency of page locking in multisystems data sharing environment.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 46% of the total text.

Page Locking Assist for Multisystems Data Sharing

      Disclosed is a technique for providing hardware lookaside
support for improving the efficiency of page locking in multisystems
data sharing environment.

      In multisystems coupled environment data sharing is often
conducted at page (e.g., 4K bytes) granule.  In such an environment,
page locking is needed to maintain data consistency cross systems.
For instance, when a system wants to update a page P (identified by
certain virtual name) it needs to acquire a page lock (P-lock) first.
The P-lock is used to guarantee that the page can be updated only by
the owner system at the moment.  During system executions a system
can be cross-interrogated for the release of a lock due to remote
system activities, in which case the system cross-interrogated will
eventually be interrupted so that the (software) lock manager can
properly release the P-lock.  This is obviously a complicated and
costly process.  As a consequence systems may tend to release P-locks
more frequently in order to reduce XI overhead.  In this invention we
provide hardware support for lock management at individual systems so
that the performance for locking can be significantly improved.
The Invention

      Consider the concept described in [*] on hardware support for
data coherence cross systems.  Each system is associated with a
lookaside table that records validity of active shared data (pages).
When a system needs to access a shared page (e.g., from its own
software buffer pool), it checks the validity of the page through its
(local) lookaside table.  Cross-system page invalidates can be
carried out (primarily through hardware) by changing relevant
lookaside tables, without interrupting individual systems.  In the
current invention similar lookaside table techniques were employed to
facilitate locking operations.

      Consider a multisystems data sharing environment in which
multiple system clusters {Si} share a common data memory space.  We
will illustrate the ideas by assuming a shared control C among all
systems (see the figure), although it is possible to generalize the
approach to distributed controls as described in [*].  C will be
responsible for resolving decisions on locking requests and
initiating XIs.  At each system Si we consider a lookaside table Li
that records information on P-locks currently held by Si.  C
communicates with the control of each individual Li via direct fast
link as described in [*].  We assume that C keeps the records on
which system is holding which lock at any moment.  Detailed locking
algorithms used by C and individual systems are not the main focus of
the invention and are dependent upon particular implementations.
Major operations relevant to the current invention are as follows:
1.   Initially Li at each system Si indicates that no P-lock is being
2.   When Si needs to know its ownership of a P-lock (e.g., for a
page P) it issues a special...