Chaudhri, V.K., Hadzilacos, V., Mylopoulos, J.: Concurrency Control for Knowledge Bases. In: Proc. of the 3rd Int. Conf. on Principles of Knowledge Representation and Reasoning, Cambridge, USA, 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of the attention of the Artificial Intelligence (AI) community, in spite of KBMSs are becoming more and more widespread and, accordingly, the demand for ever larger knowledge bases (KBs) higher and higher. Due to the ever growing applicability of KBMSs, it is time to allow for knowledge sharing [2, 3, 14]. Consequently, multiple transactions should be able to run at the same time for better performance of such systems [4] Finally, it is exactly in this point that transaction models and CC mechanisms with appropriate lock modes for KBMSs play a crucial role, because they are among the most ....

Chaudhri, V.K., Hadzilacos, V., Mylopoulos, J.: Concurrency Control for Knowledge Bases. In: Proc. of the 3rd Int. Conf. on Principles of Knowledge Representation and Reasoning, Cambridge, USA, 1992.

....on classes gives opportunities for extensibility that do not exist in the relational context. Moreover, dynamic changes in a locking graph are not a concern in [17] The only previous work we are aware of that considers a dynamically changing locking graph is in the context of a knowledge base[9]. Their solution is to restrict the order of user lock requests. Our work is in a more general 3 database context, where such restrictions on order may not be acceptable. Instead, we use intention locks. We make use of deferred updates in maintaining our lock graph. While no one has proposed the ....

Vinay Chaudhri, Vassos Hadzilacos, and John Mylopoulos. Concurrency control for knowledge bases. In Proc. 3rd Int'l Conf. on Knowledge Representation and Reasoning, October 1992.

....for KBMSs should pay attention to such types of operations, and provide adequate lock modes to cope well with them. 3 Related Work To the best of our knowledge, there is only one CC protocol tailored for KBMSs already published, namely the Dynamic Directed Graph (DDG) policy of Chaudhri et al. [CHM92]. Due to space limitations, we will not provide an exhaustive discussion about this protocol here 3 . Nevertheless, among the main drawbacks of this protocol, we can cite [Re94] First, no difference is made between different abstraction relationships, i.e. it does not treat, for example, ....

Chaudhri, V.K., Hadzilacos, V., Mylopoulos, J.: Concurrency Control for Knowledge Bases. In: Proc. 3rd KRR, Cambridge, USA, 1992.

....most of the effort has been directed at the issue of coupling the two systems and not towards an integrated architecture. The Telos KBMS approach reforms all the issues starting from the storage manager and the cost model and going to query processing and optimization [31] concurrency control [6], and integrity enforcement [28] As for future research, there exist a number of issues which demand further work. In storage management, the adoption of a physical storage kernel can give us a more precise clue of the record placement problem and the potential clustering policies. We will also ....

V. Chaudhri, V. Hadzilacos, and J. Mylopoulos. Concurrency Control for Knowledge Bases. In Proc. of KRR-92, 1992.

....case studies and instantiate the abstract directed graph abstraction to real knowledge bases. That s when the relevance and applicability of this work to knowledge bases becomes more immediate. Some of the preliminary results from Chapters 2 and 5 have already been published as conference papers (Chaudhri, Hadzilacos and Mylopoulos 1992; Chaudhri et al. 1994) A paper describing the details of theoretical results from Chapter 2 has been accepted for publication (Chaudhri and Hadzilacos 1995) A summary of the results of the thesis has been published as part of a larger paper giving system description of a knowledge base ....

Chaudhri, V. K., Hadzilacos, V., and Mylopoulos, J. 1992. Concurrency Control for Knowledge Bases. In Proceedings of the Third International Conference on Knowledge Representation and Reasoning, pages 762--773.

....can be a general graph. This characterization leads to a convenient proof technique that can be used to obtain correctness proofs for locking policies for dynamic databases. Specifically, we consider three locking policies (1) Dynamic Directed Graph policy with shared and exclusive locks (DDG SX) [CHM92, Cha94], which is an extension of the tree policy [SK80, KS83] and was designed by us for use in knowledge based applications, 2) Altruistic Locking [SGMS94] that was designed to deal with long lived transactions, and (3) Dynamic Tree Policy [CM86] that allows a changing set of partial orders over a ....

....Locking Policies for Dynamic Databases 6 and an edge represents a partof relationship between two objects. To take advantage of the graph like structure of such systems, we extended the DAG policy [SK80, Yan82] to the Dynamic Directed Graph (DDG) policy which can deal with general dynamic graphs [CHM92]. DDG SX is the variant of this policy that supports both shared and exclusive locks. We first define some properties of directed graphs that are necessary for specifying our algorithm. A root of a directed graph is a node that does not have any predecessors. A directed graph is rooted if it has ....

Vinay K Chaudhri, Vassos Hadzilacos, and John Mylopoulos. Concurrency Control for Knowledge Bases. In Proceedings of the Third International Conference on Knowledge Representation and Reasoning, pages 762--773, 1992.

....time to users [14] What is needed is a concurrency control policy that exploits the organizational and semantic structure of knowledge bases to offer efficient and correct concurrent execution of transactions. This paper examines one such policy, called the Dynamic Directed Graph (DDG) policy [6, 7]. In this paper, we study the implementation alternatives that arise as a result of knowledge base characteristics and evaluate the performance of the DDG policy in comparison to two phase locking (hereafter 2PL) 9] We choose 2PL as a yardstick in evaluating the new policy because it is the ....

....connected component as A. Thus, the root node dominates all the nodes in the graph including itself. All nodes in a strongly connected component dominate each other. With this model, we give a brief description of our locking policy in the next sub section. More details can be found elsewhere [6, 7]. 2.2 Description of the Locking Policy The Dynamic Directed Graph (DDG) policy has three types of rules: preprocessing rules, structure maintenance rules and locking rules. The preprocessing rules and the structure maintenance rules ensure that the graph is always rooted and connected; this is ....

V. K. Chaudhri, V. Hadzilacos, and J. Mylopoulos. Concurrency Control for Knowledge Bases. In Proceedings of the Third International Conference on Knowledge Representation and Reasoning, pages 762--773, 1992.

....for graph structures the abstraction of knowledge base that appears to be the most appealing. Therefore, we have adopted the locking class of methods for knowledge bases. In the rest of the paper, we will focus on locking algorithms. The discussion on the other methods can be found in (Chaudhri, Hadzilacos and Mylopoulos 1992). We will describe here two well known locking algorithms. The first, known as two phase locking, does not make any assumption about the structure of the underlying data. The second, known as the DAG policy, assumes that the underlying data is structured as a directed acyclic graph. Two phase ....

...., for i 6= 1, the prefix of T i up to and including the ( U A i ) step. S 0 is the serial execution of T 0 1 ; T 0 k in this order. 3b) Furthermore, S 0 can be extended to a complete schedule, that is, can avoid deadlock. The details of the proof of this theorem can be found in (Chaudhri, Hadzilacos and Mylopoulos 1992). Theorem 2 Dynamic Directed Graph policy is a safe policy. Proof Outline: Suppose it is not. Then, choose transactions T 1 ; T k , entities A 1 ; A k and the corresponding schedule S 0 as in Theorem 1. The serialization graph D(S 0 ) of the schedule up to (but not ....

Chaudhri, V. K., Hadzilacos, V., and Mylopoulos, J. 1992. Concurrency Control for Knowledge Bases. Technical Report Forthcoming, University of Toronto.

....transaction should acquire locks. Maintenance rules specify additional operations that must be executed by transactions to keep the structure rooted and connected. The rest of the discussion in this section focuses on locking rules. A detailed description of the DDG algorithm appears elsewhere (Chaudhri, Hadzilacos and Mylopoulos, 1992). A transaction may lock a node in shared or exclusive mode, denoted by S and X respectively (Bernstein, Hadzilacos and Goodman, 1987) The locking rules for DDG are as follows: L1. The first lock obtained by a transaction can be on any node. L2. Before a transaction T performs any INSERT, ....

....general, the DDG policy does not permit concurrency within cycles (see rule L4 above) This suggests that if a knowledge base contains very large cycles which need to be locked as one node, concurrency will be reduced. We have a version of the DDG policy that does permit concurrency within cycles (Chaudhri, Hadzilacos and Mylopoulos, 1992). We adopted the above version, because the transactions in knowledge bases tend to access all the nodes on a cycle together, and therefore, the cycles are a natural unit of locking. In order for a transaction to be able to satisfy locking rules L3a and L3b for all the nodes that it needs to lock, ....

Chaudhri, V. K., Hadzilacos, V., and Mylopoulos, J. (1992). Concurrency Control for Knowledge Bases. In Proceedings of the Third International Conference on Knowledge Representation and Reasoning, pages 762--773.

....locks; iii) maintenance rules specify additional operations that must be executed by transactions to keep the knowledge base structure in the desired form. In the sequel, we only present the locking rules of the DDG policy. A detailed description of our model and the algorithm is available in [Chaudhri92]. For the purposes of this section we will assume that the pre processing rules have been already applied to the knowledge base and the graph G is rooted and connected. In such a graph, a dominator of a set of nodes U is defined as a node d such that all paths from the root node to each node v e U ....

....lock on at least one of them. All the nodes on a strongly connected component are locked together in one step, provided all the entry points of that component have been locked. A node that is being inserted can be locked at any time. Rule L4. Each node can be locked at most once. It has been shown [Chaudhri92] that the DDG policy only generates schedules which are serializable. and deadlock free. Moreover, the DDG policy has been proven to be well structured in the sense that it will handle arbitrary transactions. This overcomes shortcomings of earlier policies that impose artificial restrictions on ....

V. Chaudhri, V. Hadzilacos and J. Mylopoulos, "Concurrency Control for Knowledge Bases", Proceedings 3rd International Conference on Principles of Knowledge Representation and Reasoning, 1992.

....does not permit concurrency within cycles (see rule L4 above) This suggests that if a knowledge base contains very large cycles which need to be locked as one node, concurrency will be reduced. We have a version of the DDG policy (called, DDG 0 policy) that does permit concurrency within cycles (Chaudhri, Hadzilacos and Mylopoulos, 1992). We adopted the above version, because the transactions in knowledge bases tend to access all the nodes on a cycle together, and therefore, the cycles are a natural unit of locking. In order for a transaction to be able to satisfy locking rules L3a and L3b for all the nodes that it needs to lock, ....

.... incremental algorithms, and therefore, the results presented in this section will scale up to larger problems The present section has presented an overview of the results based on the second author s doctoral dissertation (Chaudhri, 1995) which have also been published as conference length papers (Chaudhri, Hadzilacos and Mylopoulos, 1992; Chaudhri et al. 1994; Chaudhri and Hadzilacos, 1995; Chaudhri and Mylopoulos, 1995) 7 Integrity Constraint and Rule Management Integrity constraints specify the valid states of a knowledge base (static constraints) as well as the allowable knowledge base state transitions (dynamic ....

Chaudhri, V. K., Hadzilacos, V., and Mylopoulos, J. (1992). Concurrency Control for Knowledge Bases. In Proceedings of the Third International Conference on Knowledge Representation and Reasoning, pages 762--773.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC