In this paper, the design, implementation techniques, and initial performance evaluation of Gamma are presented. Gamma is a new relational database machine that exploits dataflow query processing techniques. Ganma is a fully operational prototype consisting of 20 VAX 11/750 computers. The design of Gamma is based on an earlier multiprocessor database machine prototype (DIRECT) and several years of subsequent research on the problems raised by the DIRECT prototype. In addition to demonstrating that parallelism can really be made to work in a database machine context, the Gamma prototype shows how parallelism can be controlled with minimal control overhead through a combination of the use of algorithms based on hashing and the pipelining of data between processes. Except for 2 messages to initiate each operator of a query tree and 1 message when the operator terminates, the execution of a query is entirely self-scheduling. 52 refs., 12 figs.

/pdf/gamma-a-high-performance-dataflow-database-machine-memstoqz39.pdf

GAMMA—a high performance dataflow database machine

In this paper we present a new algorithm, DBMIN, for managing the buffer pool of a relational database manaegememt system. DBMIN is based on a new model of relational query behavior, the query locality set model (QLSM). Like the hot set model, the QLSM has an advantage over the stochastic models due to its ability to predict future reference hehavior. However, the QLSM avoids the potential problems of the hot set model by separating the modeling of referr-ence bahavior from any particular buffer management algorithm. After introducing the QLSM and describing the DBMIN algorithm, we present a performance evaluation methodology for evaluating buffer manage-ment algorithms in a multiuser environment. This methodology employed a hybrid model that comhines features of both trace driven and distribution driven simulation models. Using this model the performance of the DBMIN algorithm in a multiuser environment is compared with that of the hot set algorithm and four more traditional buffer replacement algorithms.

/pdf/an-evaluation-of-buffer-management-strategies-for-relational-59eqiec1mr.pdf

An evaluation of buffer management strategies for relational database systems

Few database query optimizer models have been validated against actual performance. This paper extends an earlier optimizer validation and performance evaluation of R’ to di.rfribu& queries, i.e. single SQL statements having tables at multiple sites. Actual R* message, I/O, and CPU resources consumed and the corresponding costs estimated by the optimizer were written to database tables using new SQL commands, permitting automated control from application programs for collecting, reducing, and comparing test data. A number of tests were rnn over a wide variety of dynamically-created test databases, SQL queries, and system parameters. Both high-speed networks (comparable to a local area network) and medium-speed long-haul networks (for linking geographically dispersed hosts) were evaluated. The tests confirmed the accuracy of R*‘s message cost model and the significant contribution of local (CPU and I/O) costs, even for a medium-speed network. Although distributed queries consume more resources overall, the response time for some execution strategies improves disproportionately by exploiting both concurrency and reduced contention for buffers. For distributed joins in which a copy of the inner table must be transferred to the join site, shipping the whole inner table dominated the strategy of fetching only those inner tuples that matched each outer-table value, even though the former strategy may require additional I/O. Bloomjoins (hashed semijoins) consistently performed better than semijoins and the best R* strategies. Few of the distributed optimizer models proposed over the last decade CAPER 83, BERN 81~. CHAN 82, CHU 82, EPST 78, HEVN 79. KERS 82, ONUE 83, PERR 84, WONG 83, YAO 79, YU 831 have been validated by comparison with actual performance. The only known validations, for Distributed INGRES [STON 821 and the Crystal multicomputer [LU 853, have assumed only a high-speed local-area network linking the distributed systems. Also, the Distributed INGRES study focused primarily on reducing response time by exploiting parallelism using table partitioning and broadcast messages. In contrast, R* seeks to minimize total resonrces consumed, has not implemented table partitionings, and does not presume a network broadcast capability. There are many important questions that a thorough validation should answer: . Under what circumstances (regions of the parameter space) does the optimizer choose a suboptimal plan, or, worse, a particularly bad plan? . To which parameters is the actual performance most sensitive? . Are these parameters being modeled accurately by the optimizer? . What is the impact of variations from the optimizer’s simplifying as. Is it possible to simplify the optimizer’s model (by using heuristics, for

R* Optimizer Validation and Performance Evaluation for Distributed Queries

On presente plusieurs techniques d'optimisation des questions au sein d'un reseau de bases de donnees. Les caracteristiques d'un algorithme integrant ces techniques sont enoncees

/pdf/distributed-query-processing-2vumc93jfj.pdf

Distributed query processing

The quantitative evaluation of computer performance is needed during the entire life cycle of a computer system. We survey the major quantitative methods used in computer performance evaluation, focusing on post-1970 developments and emphasizing trends and challenges. We divide the methods used into three main areas, namely performance measurement, analytic performance modeling, and simulation performance modeling, which we survey in the three main sections of the paper. Although we concentrate on the methods per se, rather than on the results of applying the methods, numerous application examples are cited. The methods to be covered have been applied across the entire spectrum of computer systems from personal computers to large mainframes and supercomputers, including both centralized and distributed systems. The application of these methods has certainly not decreased over the years and we anticipate their continued use as well as their enhancement when needed to evaluate future systems.

Computer Performance Evaluation Methodology

In this paper we examine four performance enhancements to a database management system: dynamic compilation, microcoded routines, a special-purpose file system, and a special-purpose operating system. All were examined in the context of the INGRES database management system. Benchmark timings that are included suggest the attractiveness of dynamic compilation and a special-purpose file system. Microcode and a special-purpose operating system are analyzed and appear to be of more limited utility in the INGRES context.

Performance enhancements to a relational database system

Performance Analysis of Distributed Data Base Systems

Performance Analysis of Distributed Data Base Systems.

In this paper we briefly present the design of a distributed relational data base system. Then, we discuss experimental observations of the performance of that system executing both short and long commands. Conclusions are also drawn concerning metrics that distributed query processing heuristics should attempt to minimize. Lastly, we comment on architectures which appear viable for distributed data base applications.

Jeff Ranstrom

Papers

Performance enhancements to a relational database system

Performance Analysis of Distributed Data Base Systems

Performance Analysis of Distributed Data Base Systems.

Performance enhancements to a relational database system

Performance Analysis of Distributed Data Base Systems.