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

Optimistic Concurrency Control With Adaptive Control of Abort Probability

IP.com Disclosure Number: IPCOM000099926D
Original Publication Date: 1990-Mar-01
Included in the Prior Art Database: 2005-Mar-15
Document File: 2 page(s) / 82K

Publishing Venue

IBM

Related People

Dias, DM: AUTHOR [+2]

Abstract

Disclosed is a concurrency control method for a transaction processing system that optimizes performance by striking a balance between aborting transactions and waiting for locks. The Optimistic type of Concurrency Control (OCC) schemes can suffer from resource limitations due to resources consumed in reruns, while the standard locking scheme can be negatively affected by the lock wait phenomenon. A method is described to alleviate the number of reruns in the OCC protocol by resorting to locking at the latter stages of the transaction execution. This is referred to as OCC with adaptive control of abort probability. By using the disclosed method, significantly better performance can be obtained at high levels of data contention.

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

Optimistic Concurrency Control With Adaptive Control of Abort Probability

       Disclosed is a concurrency control method for a
transaction processing system that optimizes performance by striking
a balance between aborting transactions and waiting for locks. The
Optimistic type of Concurrency Control (OCC) schemes can suffer from
resource limitations due to resources consumed in reruns, while the
standard locking scheme can be negatively affected by the lock wait
phenomenon. A method is described to alleviate the number of reruns
in the OCC protocol by resorting to locking at the latter stages of
the transaction execution.  This is referred to as OCC with adaptive
control of abort probability.  By using the disclosed method,
significantly better performance can be obtained at high levels of
data contention.

      Let N be the average number of granules accessed by each
transaction. Assume that from trace analysis or online measurement we
can obtain an estimate of N. Define K to be a critical stage of
transaction execution to be determined later. At the beginning of the
transaction execution, the disclosed method operates similar to the
pure OCC scheme (1,2). However, after a transaction has accessed K
granules, the following two cases occur. (a) If the transaction has
not yet been marked for abort, and if a lock can be obtained for each
of the K granules accessed, then locks on these K granules are
obtained; subsequently, lock requests are made for each new granule
accessed and the transaction will wait for a requested lock if the
granule is held by another transaction in incompatible mode. (b) If
the transaction has already been marked for abort or if locks on all
the K granules, accesses thus far cannot be obtained, the transaction
is marked for abort and execution continues in optimistic mode. At
the end of the transaction (commit time) if the transaction has been
marked for abort it is rerun; otherwise, the transaction is committed
and any r...