Dynamická optimalizace ukládání hybridních self-referenčních sémantických sítí

Bc. Miroslav Kintr

Master's thesis

Dynamic storage optimization for hybrid self-referential semantic networks

Abstract:

To balance highest available performance of query execution and a reasonable amount of structures necessary to make that performance possible is a difficult task in most database applications. By performance of query execution we mean how quick we get the answer to a posted query. The faster the execution shall be the more redundant data, e.g. more indexes, we have to store. The amount of memory needed to store such redundant data will be called the cost of the query execution performance. Usually the designer of the database application can figure out the most usual queries and design the appropriate structures according to it. But there are applications he can not. This is the case of a hybrid self-referential semantic network for example. In general at design time we can not tell which queries will be the most frequent and which will be rather unusual. Another complication is introduced by the fact that the database of binary connections, which are the atomic pieces of our hybrid self-referential semantic network, is stored in a hierarchical DBMS for performance reasons instead of a relational DBMS with its declarative data definition and querying languages. In this work we will introduce our approach to solving the above-mentioned optimization problem in a hierarchical database. Three steps are necessary to be done. The first is a database API generator. A tool containing a database of facts about the logical and physical view of structures stored in the database and rules how to generate program code managing these structures. The second is an algorithm for choosing among available structures the optimal one to query. The third is just the optimization algorithm for choosing a subset from valid bamboo global set where synapses of a given type will be stored. The dynamics of this optimization algorithm is in its ability to choose structures according to usage frequency of these structures collected in the past. Via one parameter the algorithm can be adjusted to focus more on performance (i.e. more globals -- more query variants executed effectively) or cost reduction (i.e. less globals -- less storage space used). Finally we will introduce the vision of a tool able to self-optimize its storage without human interaction and mention the basics of its design.

Keywords

semantic network, code generator, relational database, hierarchical database, synapse oriented approach, storage optimization, bamboo, synapse, global

Language used: English

Date on which the thesis was submitted / produced: 5. 1. 2007

Thesis defence

• Date of defence: 12. 2. 2007
• Supervisor: RNDr. Zdenko Staníček, Ph.D.

Masaryk University

Faculty of Informatics

Master programme / field:
Applied Informatics / Applied Informatics