Kamel Boukhalfa

SAGA: a combination of genetic and simulated annealing algorithms for physical data warehouse design

Data partitioning is one of the physical data warehouse design techniques that accelerates OLAP queries and facilitates the warehouse manageability. To partition a relational warehouse, the best way consists in fragmenting dimension tables and then using their fragmentation schemas to partition the fact table. This type of fragmentation may dramatically increase the number of fragments of the fact table and makes their maintenance very costly. However, the search space for selecting an optimal fragmentation schema in the data warehouse context may be exponentially large. In this paper, the horizontal fragmentation selection problem is formalised as an optimisation problem with a maintenance constraint representing the number of fragments that the data warehouse administrator may manage. To deal with this problem, we present, SAGA, a hybrid method combining a genetic and a simulated annealing algorithms. We conduct several experimental studies using the APB-1 release II benchmark in order to validate our proposed algorithms.

Data Partitioning in Data Warehouses: Hardness Study, Heuristics and ORACLE Validation

Horizontal data partitioning is a non redundant optimization technique used in designing data warehouses. Most of todays commercial database systems offer native data definition language support for defining horizontal partitions of a table. Two types of horizontal partitioning are available: primary and derived horizontal fragmentations. In the first type, a table is decomposed into a set of fragments based on its own attributes, whereas in the second type, a table is fragmented based on partitioning schemes of other tables. In this paper, we first show hardness to select an optimal partitioning schema of a relational data warehouse. Due to its high complexity, we develop a hill climbing algorithm to select a near optimal solution. Finally, we conduct extensive experimental studies to compare the proposed algorithm with the existing ones using a mathematical cost model. The generated fragmentation schemes by these algorithms are validated on Oracle 10g using data set of APB1 benchmark.

Yet another algorithms for selecting bitmap join indexes

One of the fundamental tasks that data warehouse (DW) administrator needs to perform during the physical design is to select the right indexes to speed up her/his queries. Two categories of indexes are available and supported by the main DBMS vendors: (i) indexes defined on a single table and (ii) indexes defined on multiple tables such as join indexes, bitmap join indexes, etc. Selecting relevant indexes for a given workload is a NP-hard problem. A majority of studies on index selection problem was focused on single table indexes, where several types of algorithms were proposed: greedy search, genetic, linear programming, etc. Parallel to these research efforts, commercial DBMS gave the same attention to single table indexes, where automated tools and advisors generating recommended indexes for a particular workload and constraints are developed. Unfortunately, only few studies dealing with the problem of selecting bitmap join indexes are carried out. Due to the high complexity of this problem, these studies mainly focused on proposing pruning solutions of the search space by the means of data mining techniques. The lack of bitmap join index selection algorithms motivates our proposal. This paper proposes selection strategies for single and multiple attributes BJI. Intensive experiments are conducted comparing the proposed strategies using mathematical cost model and the obtained results are validated under Oracle using APB1 benchmark.

Pruning search space of physical database design

Very large databases and data warehouses require many optimization structures to speed up their queries. These structures can be classified into two main categories: (1) redundant structures like mono attribute indexes, multi-attribute indexes (bitmap join indexes), materialized views, etc. and (2) no redundant structures, like horizontal partitioning and vertical partitioning. The problem of selecting any of these structures is a very crucial decision for the performance of the data warehouse. In this work, we focus on horizontal partitioning and bitmap join indexes. We first show the similarity between horizontal partitioning and bitmap join indexes. Secondly, we propose a new approach of selecting simultaneously these structures in order to reduce the query processing cost. It consists in using the horizontal partitioning schema obtained by a genetic algorithm to prune the search space of the problem of bitmap join index selection. Thirdly, we propose a greedy algorithm to select bitmap join indexes under a storage bound. Finally, we conduct several experimental studies using an adaptation of APB-1 benchmark in order to validate our proposed algorithms.

SimulPh.D.: A Physical Design Simulator Tool

The importance of physical design has been amplified as query optimizers became sophisticated to cope with complex decision support applications. During the physical design phase, the database designer (DBD) has to select optimization techniques to improve query performance and to well manage different resources assigned for his/her databases. The decision of using these optimization techniques is taken either before or after creating the database schema. Once this decision taken, DBD has to perform three main tasks: (i) choosing one or several optimization techniques, (ii) managing interdependencies among the chosen techniques and (iii) choosing a selection algorithm for each technique. Faced to these crucial choices, the development of simulators intended to improve the quality of the physical design represents challenging issue. In this paper, we propose a simulator, called SimulPh.D that assists DBD to perform different choices thanks to user friendly graphical interfaces.

A Cost Model for DBaaS Storage

Cloud infrastructures employ hybrid storage systems that incorporate various types of devices flash memory solid-state and hard disk drives. Dealing with such heterogeneity makes the use of data placements strategies necessary. These strategies generally rely on cost modeling techniques. In this paper, we propose a cost model for the storage of database objects in a Cloud infrastructure. Our cost model increments the existing work by including: 1 storage cost, which comprises the occupation, the energy and the endurance costs, 2 the penalty cost that could arise from the SLA Service Level Agreement violation, and 3 the migration cost resulting from the object movement between storage systems. We also evaluate the relevance of our model and its usability throughout examples.

SOLAP On-the-Fly Generalization Approach Based on Spatial Hierarchical Structures

On-the-fly generalization, denotes the use of automated generalization techniques in real-time. This process creates a temporary, generalized dataset exclusively for visualization, not for storage or other purposes. This makes the process well suited to highly interactive applications such as online mapping, mobile mapping and SOLAP. BLG tree is a spatial hierarchical structure widely used in cartographic map generalization and particularly in the context of web mapping. However, this structure is insufficient in the context of SOLAP applications, because it is mainly dedicated to the geographic information processing (geometric features), while SOLAP applications manage a very important decision information that is the measure. In this paper, we propose a new structure, SOLAP BLG Tree, adapted to the generalizaion process in the SOLAP context. Our generalization approach is based on this structure and uses the simplification operator. Combining the topological aspect of geographical objects and the decisional aspect (the measure).

De l'optimisation de requêtes aux outils d'administration des entrepôts de données

Administrating a data warehouse becomes a crucial issue compare to traditional databases. During the administration phase, data warehouse administrator shall establish several choices. These choices may concern optimisation techniques, selection algorithms of these techniques and their parameters, attributes and tables used by some techniques (such as indexing and partitioning). Note that optimization techniques evolve since data warehouse dynamically changes during its lifetime. These changes necessitate a tuning phase so as to keep the performance ofwarehouse from degrading. In this paper, we firstly, show difficulties that administrator has during her/his administration and tuning tasks. Secondly, we propose a tuning method based on horizontal partitioning and bitmap join indexes. Finally, we present a tool advising the administrator in an interactive way to realise his/her tasks.

Design and Implementation of Falling Star - A Non-Redudant Spatio-Multidimensional Logical Model for Document Stores.

In the context of Spatial Big Data, some NoSQL spatial DBMSs have been developed to deal with the Spatiality, Velocity, Variety, and Volume of Spatial Big Data. In this context, some works recently study NoSQL logical Data Warehouse (DW) models. However, these proposals do not investigate storing and querying spatial data. Therefore, in this paper we, propose a new logical model for document Spatial DWs. Moreover, motivated by the expressivity, readability and interoperability offered by UML profile, we represent our model using a UML profile. Finally, we present an implementation in document Spatial DBMSs.

COPS: Cost Based Object Placement Strategies on Hybrid Storage System for DBaaS Cloud

Solid State Drives (SSD) are integrated together with Hard Disk Drives (HDD) in Hybrid Storage Systems (HSS) for Cloud environment. When it comes to storing data, some placement strategies are used to find the best location (SSD or HDD). These strategies should minimize the cost of data placement while satisfying Service Level Objectives (SLO). This paper presents two Cost based Object Placement Strategies (COPS) for DBaaS objects in HSS: a Genetic based approach (G-COPS) and an ad-hoc Heuristic approach (H-COPS) based on incremental optimization. While G-COPS proved to be closer to the optimal solution in case of small instances, H-COPS showed a better scalability as it approached the exact solution even for large instances (by 10% in average). In addition, H-COPS showed small execution times (few seconds) even for large instances which makes it a good candidate to be used in runtime. Both H-COPS and G-COPS performed better than state-of-the-art solutions as they satisfied SLOs while reducing the overall cost by more than 40% for problems of small and large instances.