| Abstract | A distributed object-oriented database system model is proposed. In this model, a
collection of objects distributed over multiple nodes which are interconnected by a
network. The network provides sharing of information between these nodes. The proposed
model differs from the traditional models in a major aspect, which is the proposed model
incorporates the concept of Abstract Data Type. That is, the database information is stored
in typed objects ( i.e. instances of abstract data type) and manipulated only by operations that
are specific to a particular object type.
In this distributed object-oriented database system model, the distribution of data
is invisible to users, users see the system as a single integrated database. Since users may
issue transactions from their local sites, many transactions are allowed to be active
simultaneously in the system. Therefore, a concurrency control scheme is required.
Based on the proposed model, a type-specific multiversion timestamping
concurrency control algorithm which is employed to synchronize transaction operations on
shared objects is developed. We give a formal description of the algorithm, encompassing
both concurrency control and recovery strategy, and prove that the algorithm satisfies ยท
Static Atomicity, a Local Atomicity Propery that ensures global atomicity in distributed
object-oriented database system. This algorithm has the following novel aspects. First,
unlike many methods that classify operations only as reads and writes, this technique
exploit the semantics of abstract data types to achieve greater concurrency. Synchronization
is based on a predefined compatibility table. Second, this technique supports modular
design. Global Atomicity of the system is achieved by satisfying properties that are local
to individual objects. Third, this technique is more integrated, it analyzes serializability and
recovery together, rather than treating them separately.
A performance evaluation is carried out to compare the level of concurrency offered
when we use the proposed concurrency control algorithm as opposed to using traditional
scheme without exploiting semantic information. The evaluation result is that the proposed
algorithm is suitable for distributed object-oriented databases and can achieve better
concurrency. |