Marie-Anne Neimat

High-availability LH* schemes with mirroring

Mirroring is a popular technique for enhancing file availability. The authors incorporate this technique into the LH* algorithms for scalable distributed linear hash files. Several schemes for mirroring LH* files are presented in this paper. The schemes increase the availability of LH* files in the presence of node failures. Every record remains accessible in the presence of a single node failure, and usually in the presence of multiple-node failures. The price is, as usual, twice as much storage for data, and an increase in the number of messages. The different schemes are characterized by different trade-offs, and they accommodate diverse application requirements. The additional messaging cost per insert is about the same for all the schemes, and is roughly only one message. The cost of a bucket recovery may in contrast vary greatly, from one message for one type of scheme, to a few for another, and many for yet another.

k-RP*s: a scalable distributed data structure for high-performance multi-attribute access

k-RP*s is a new data structure for scalable multicomputer files with multi-attribute (k-d) keys. We discuss the k-RP*s file evolution and search algorithms. Performance analysis shows that a k-RP*s file can be much larger and orders of magnitude faster than a traditional k-d file. The speed-up is especially important for range and partial match searches that are often impractical with traditional k-d files. This opens up a new perspective for many applications.

LH*s: a high-availability and high-security scalable distributed data structure

LH*s is high availability variant of LH*, a Scalable Distributed Data Structure. An LH*s record is striped onto different server nodes. A parity segment allows one to reconstruct the record if a segment fails. The insert or key search time is about a msec on a 10 Mb/s net, and about 100 /spl mu/s at 1 Gb/s net, assuming the segments in the distributed RAM. The file size depends only on the distributed storage available, i.e., a RAM file can reach dozens of GB in practice. Data security is enhanced, as every site contains only partial and typically meaningless data. The price to pay is 20-50% more storage for the file than for an LH* file, and some additional messaging, especially for the scan search.

The papyrus integrated data server

Summary form only given. The authors focus on the performance of integrated specialized data managers. In particular, they focus on the customizations of parallel executions of data manager operators in a variety of computer configurations. This is done by specifying the glue that connects data manager operators in a way that is independent of the computer configuration; and then providing the ability to transparently target the execution to a variety of computer configurations. Parallelization of data manager operators built using Papyrus services is a challenging problem, but a more challenging problem is the parallelization of data manager operators that are built independently of Papyrus and are therefore black boxes to Papyrus. Papyrus is a set of modules and services that enables the parallelization and integration of specialized data managers into one execution environment. Papyrus programs can be transparently targeted to different hardware configurations and can dynamically adjust at runtime to the number of available resources. A Papyrus System consists of a number of clients interfacing to a Papyrus Server. The Server consists of several integrated data managers executing on a multiprocessor system.<<ETX>>

Optimization of relational algebra expressions containing recursion operators

Efficient computation of recursive queries is one of the key issues in the development of next generation database management systems. In this paper, we extend the relational algebra with a fixpoint operator that supports the definition of recursive relations. Legal transformation rules on relational algebra expressions with the fixpoint operator are then investigated. They are used to convert relational algebra expressions to equivalent expressions that can be more efficiently evaluated. 1 I n t r o d u c t i o n Recently, a great deal of effort has been devoted to extend the functionality of existing database management systems (DBMSs). Recursion is one of the most desirable extensions. It is well known that conventional relational algebra is not sufficient for processing recursive queries [2]. New operators, such as transitive closure operators o~ [1] and o ~ [4], have been proposed to extend the power of relational algebra. Ilowever, we feel that query processing for recursive queries should be developed in a more general context. It should be able to handle general linear recursion as classified in [19] as well as mutual recursion, because many recursive queries cannot be translated into transitive closure forms. Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specific permission. The efficient processing of recursive queries is one of the major technical issues facing the developers of new generation DBMSs. A number of strategies and algorithms have been proposed in the literature and Bancilhon and Ramakrishnan have a comprehensive survey on this problem [3]. Some of these efforts consider the high-level refinement of logic expressions [4, 7, 10, 17, 20], others the low-level efficient computation of transitive closure queries[8, 9, 11, 18]. To benefit from the well developcd relational technology, we would like to integrate recursive query processing into conventional query processing. In our view, the successful support of recursive queries requires that their evaluation be tightly coupled with query optimization and execution techniques used in current relational systems [15, 16]. This includes: 1. the design of a relational algebra operator powerful enough to express the semantics of recursive queries, 2. the development of optimization strategies for relational expressions involving classical relational algebra operators and the recursire operator, and 3. the development of efficient algorithms and data structures for implementing the recursire operator. To achieve our goal, we extend the relational algebra with a fixpoint operator that can define linear recursive relations as well as nmtually recursive relations. We then investigate relational algebraic laws to transform relational expressions that include fixpoint operators into equivalent expressions thus giving the query optimizer more flexi© 1991 ACM 089791-382-5/91/0003/0332

Cost Model Development for a Main Memory Database System

1.50 332 bility in choosing the best execution strategy. The implementation of the fixpoint operator has been explored in [14]. The contributions of this paper are in the definition of an operator that can handle mutual recursion and in exploring the transformation rules applicable to this operator when it is embedded in relational algebra expressions. To the best of our knowledge, all proposed extensions to relational algebra have been with operators that can only handle transitive closure. In addition, the work on optimizing recursive expressions has been very specialized such as the work in [6], which explores transformation rules on a transitive closure operator with the selection operator. The remainder of this paper is organized as follows: Section 2 proposes the fixpoint operator as a powerful addition to traditional relational algebra. Section 3 presents some examples with the fixpoint operator. Section 4 is the heart of this paper. The relationship between the fixpoint operator and other relational operators is investigated. Finally, in Section 5, we summarize the main conclusions of this study. 2 F i x p o i n t o p e r a t o r In order to take advantage of query optimization techniques developed in relational systems, we choose to extend the relational algebra rather than develop a completely new theoretic foundation. A powerful operator ®, called fixpoint operator, is introduced into the relational algebra. Its purpose is to enhance the declarative power of conventional relational algebra by supporting recursive queries. As with all other relational operators, the operands and output of the fixpoint operator ® are relations. That is, its input is one or more relations and its output is a single relation. In addition, as its name implies, the fixpoint operator supports the least fixed point semantics. Previous proposals, such as the a operator [1], and the o x operator [4], have been limited to support the transitive Closure of a single relation. They take a single relation for input and produce its transitive closure for output . Our proposed fixpoint operator ® can compute linear recursion and, more importantly, can compute mutually recursive relations. V~re first present the basic forms of this new operator and then use them as basic building blocks to compose more genera] recursive queries. The simple form of the ® operator is defined as follows:

Interoperability of Spatial and Attribute Data Managers: A Case Study

Main-memory database management systems (MM-DBMS’s) are at the heart of RTDB’s, and research in MM-DBMS’s has been active since the mid-eighties [8, 7, 1, 9]. Recently the interest has taken on a new urgency as inexpensive memory and 64-bit addressing are becoming reality. Several fairly complete systems [12, 14, 3] have been developed in the last few years, and recent investigations have taken a fresh look at a variety of issues in the context of main-memory: recovery [15, 19, 20, 16], indexing [2, 23], parallelism [3], and concurrency control [11], for example. However, the issue of query optimization has largely been neglected, partly because many of the applications suited to main-memory systems (e.g., telecom switching and financial trading) use only simple queries requiring, say, a hash lookup on a single table. There are, however, a few applications that require complex queries over memory-resident data. These include financial analysis, and fraud detection in the context of telecommunication. Moreover, we consider main-memory databases to be a “disruptive technology” [5] and so we anticipate that as the technology becomes more widely adopted, MM-DBMS’s will be used in increasingly general-purpose situations, which will require query optimization. Indeed, the recent announcement that Oracle will be including a copy of an in-memory database with each Oracle7 system [22] goes some way toward justifying this belief. Similarly, the emerging popularity of object-oriented DBMS’s, which is partly due to their high performance, is to a great extent attributable to the memory residence of the data.

The Papyrus Object Library

We relate in this paper our experience in integrating a commercial relational storage manager and a commercial spatial data manager. We present the challenge of interoperating such data managers and expose the problems that must be solved to make such interoperability feasible, painless, transparent, and efficient. We present the approach we have taken in the Papyrus system at solving some of these problems, and discuss the issues that must be addressed to solve the remaining problems. Through examples, we emphasize the need for a comprehensive cost-based query optimization strategy, and show that the lack of such a strategy can result in a system with unacceptable performance.

Database research at HP labs

The Papyrus Object Library is a set of routines that provide simple access to persistent recoverable storage. It is intended to be used as one of several tools for implementing Data Managers in Papyrus. Three primary goals have shaped the design of the Object Library. The first is flexibility in adapting to Data Manager needs. The Object Library can be used to implement traditional storage managers as well as to provide persistence to programming languages. The second goal is to insulate Data Manager implementors from operating system and file system details without sacrificing performance. Simple localized modifications to one module of the Library can be easily made to take advantage of operating system features that can improve performance such as mapped files or raw I/O. The last goal is to provide high performance. The system is designed to provide very fast access to objects and very efficient allocation and deallocation of objects.

Achieving transaction scaleup on Unix

Database research at HP Labs has focused in the last few years on the productive application development and use of database systems. The approach centered on developing a rich database model for expressing the semantics and behavior of information. The Iris object-oriented database management system is the result of that e ort. Iris combines the advantages of object-oriented concepts with database functionality. The advent of open systems and fast networks has fostered the belief that information is available at ones ngertips. Yet these advances have pointed to heterogeneity as the most serious challenge to programmer and knowledge worker productivity. Heterogeneous database systems must resolve the semantic discrepancies between di erent databases, handle the conversions between multiple data models, schemas and formats, etc., while preserving the autonomy of the underlying systems. Preserving the autonomy of the component databases is essential because they are supplied by di erent vendors or are large systems that took many years to develop, which makes modifying them either impossible or impractical. There is another class of data management systems that database researchers have not attempted to integrate, and yet these systems manage the bulk of data currently stored on computer systems. They are the specialized database systems such as spatial, geographical and CAD/CAM DBMSs. These customized systems were developed and optimized to meet the performance requirements of their respective applications. A new challenge is to interoperate these heterogeneous database systems while preserving their performance requirements. In integrating this class of systems, preserving their autonomy is no longer the primary requirement, but rather preserving the performance of the integrated system. Exploiting parallel and distributed execution is essential for meeting/exceeding the performance requirements for the integrated systems. We view autonomy and performance as two requirements of heterogenous databases at two ends of a spectrum where one must be traded to obtain the other. This report describes the highlights of the Iris work and the current database research e orts in tradi-