Skip to main content
SpringerLink
Log in
Book cover

Database Machine Performance: Modeling Methodologies and Evaluation Strategies pp 208–250Cite as

Chapter 10 Database machine design and performance evaluation: Annotated bibliography

  • Chapter
  • First Online:

  • 181 Accesses

Part of the book series: Lecture Notes in Computer Science ((LNCS,volume 257))

This is a preview of subscription content, log in via an institution.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

10.3 — Annotated References

  1. A.Avizienis, A.F.Cardenas, F.Alavian: On the Effectiveness of Fault-Tolerance Techniques in Parallel Associative Database Processors, Proc. of Int. Conf. on Data Engineering, Los Angeles, 1984, pp. 50–59. The paper deals with some fault-tolerant techniques which differ from those proposed in [CAA83]. A general model representing machines like CASSM and RAP is described and fault-tolerance is systematically applied to its organization. Storage areas are protected by duplication and error detecting and/or correcting codes. The area processors, which search storage areas, are replicated and periodically checked. The effectiveness of these techniques is shown by the analytic results obtained by a program based on a unified Markov reliability model.

    Google Scholar 

  2. R. Agrawal and D. J. DeWitt: Recovery Architecture for Multiprocessor Database Machines, ACM SIGMOD, Austin, Texas, 1985, pp. 131–145. This study deals with recovery and its impact on performance of database machines. The authors propose several parallel recovery architectures for multiprocessor database machines and examine their characteristics in detail; they then evaluate the impact of the results on database machine performance. Log, shadows and differential files mechanisms are examined and performance is evaluated by means of simulation experiments. Two metrics are used for studying the performance: average execution time per page and average transaction completion time. The results indicate that a recovery architecture based on parallel logging has the best overall performance.

    Google Scholar 

  3. R. Agrawal: A Parallel Logging Algorithm for Multiprocessor Database Machines, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, 1985, Springer-Verlag, 1985, pp. 256–276. In this paper a recovery architecture based on parallel logging for multiprocessor-cache database machines is presented. Simulation experiments are made in order to determine the characteristics of the parallel logging algorithm and its impact on database machine performance. The author shows how the recovery actions can be completely overlapped with the data page processing so that the performance of the database machine is not degraded by recovery overhead.

    Google Scholar 

  4. Amperif Corporation: The Relational Database Machine RDM-1100, Amperif Corporation, 1982, Chatsworth, California. The RDM 1100 was introduced for use with UNIVAC 1100 host computers and Amperif disk drives; it uses the Britton-Lee IDM (see [BRI81]) internally under its cover. Interface software which permits the host computer to utilize the RDM 1100 requires no modification in the UNIVAC operating system. A relational query language provides a high-level, on-line interface to the RDM 1100.

    Google Scholar 

  5. J.P.Armisen, J.Y.Caleca: A commercial back-end data base system, Proc. of 7th Int. Conf. on Very Large Data Bases, Cannes, 1981, pp. 56–65. This paper describes the M1X database machine developed for commercialization. The back-end supports both a Codasyl and a relational interface, and locking and recovery mechanisms are included.

    Google Scholar 

  6. H.Auer, H.Ch.Zeidler: On the Development of Dedicated Hardware for Searching, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 346–365. The central points of this paper regard the features required for implementing the logic necessary for forming a search processor. The authors discuss the design of data filters and more in particular they examine the search processor of the RDBM relational database machine (see [SZH83]).

    Google Scholar 

  7. E. Babb: Implementing a Relational Database by means of Specialized Hardware, ACM TODS, vol. 4, n. 1, March 1979, pp. 1–29. This paper deals with CAFS which is a special-purpose peripheral device designed for handling database transactions in a multiuser environment. In order to perform projection and join, the CAFS system possesses a random access store which contains an array of single-bit elements addressable by the key-field in a tuple. A method of addressing the bit array store by using hashing techniques is given and algorithms using the hashed bit array store to perform join and projection are described. Theoretical and experimental results regarding the behaviour of the hashed single-bit array store are also given.

    Google Scholar 

  8. D. Bitton, H. Boral, D.J. DeWitt and W.K. Wilkinson: Parallel Algorithms for the Execution of Relational Database Operations, ACM TODS, Vol. 8, n.3, Sept. 1983, pp. 324–353. Parallel algorithms for sorting, projection and join operations in a generalized multiprocessor environment are presented and analyzed by means of deterministic models. Cases in which the number of pages is significantly larger than the number of processors are examined.

    Google Scholar 

  9. Bogdanowicz, M.Crocker, D.K.Hsiao, C.Ryder, V.Stone and P.Strawser: Experiments in Benchmarking Relational Database Machines, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 106–134. A description of a large set of benchmarking experiments on a relational database machine is presented. The experiments are based on artificial databases which are defined by means of a database generation tool. The benchmark query set includes select, projection, and join operations. One query at a time is run in the machine. The definition of a machine-independent methodology for benchmarking database machines is an important result of this study.

    Google Scholar 

  10. D. Bitton, D.J. DeWitt, D.K. Hsiao, J. Menon: A Taxonomy of Parallel Sorting, ACM Computing Surveys, vol.16, n.3, September 1984, pp. 287–318. Parallel sorting algorithms are discussed according to several criteria which refer both to the time complexity of the algorithms and their architectural requirements.

    Google Scholar 

  11. D.Bitton, D.J.DeWitt and C.Turbyfill: Benchmarking Database Systems A Systematic Approach, Proc. of 9th Conf. on VLDB, Florence, 1983, pp. 8–19. Commercial and university INGRES database system versions and the IDM-500 database machine are compared together by using a simple but carefully tuned relational database. A comprehensive set of queries, such as selection, join, projection, aggregate, and update is also used. The benchmarking does not take a multiuser environment into account.

    Google Scholar 

  12. D.Bitton, D.J.DeWitt and C.Turbyfill: Benchmarking Database Systems a Systematic Approach, Computer Science Department Technical Report, n. 526, Univ. of Wisconsin, October 1983. This report is a revised and expanded version of the [BDT83a] paper in which the ORACLE database system and the DIRECT database machine are added to the original comparison.

    Google Scholar 

  13. H. Boral, D.J. DeWitt and W.K. Wilkinson: Performance Evaluation of Four Associative Disk Designs, Information Systems, Vol. 7, n.1, 1982, pp.53–64. The results of an event-driven simulation of associative disk architectures are presented. The PPH (Processor-Per-Head), PPT (Processor-Per-Track), PPB (Processor-Per-Bubble-Cell), and PPD (Processor-Per-Disk) designs are analyzed.

    Google Scholar 

  14. P.B. Berra, E. Oliver: The Role of Associative Array Processors in Data Base Machine Architecture, Computer, vol.12, n.3, 1979, pp. 53–61. This paper discusses the utilization of the STARAN associative array processor in data base management. This processor was built by the Goodyear Aerospace Corporation and was originally designed for image processing. It includes an array, which is the storage device containing the data, a comparand register which contains the argument, a mask register which determines whether or not the bit slices of the array are to function when a given operation has to be performed, response registers which record search results, perform boolean operations and provide word selection capability. Three general configurations are discussed and in all three the data are staged from the auxiliary memory into the associative array processor used for searching, retrieving and updating a large data base.

    Google Scholar 

  15. F.Bancilhon, D.Fortin, S.Gamerman, J.M.Laubin, P.Richard, M.Scholl, D.Tusera, A.Verroust: VERSO — A Relational Backend Database Machine, in "Advanced Database Machine Architecture", D.K.Hsiao ed., Prentice-Hall, 1983, pp. 1–18. The VERSO database machine is built around a fast filtering device designed for relational databases and based on a programmable finite state automaton. Its filter is capable of executing unary operations on-the-fly and its binary operation execution requires sorted data.

    Google Scholar 

  16. J. Banerjee, D.K. Hsiao and R. Baum: Concepts and Capabitilies of a Database Computer, ACM TODS, Vol. 3, n.4, Dec. 1978, pp. 347–384. This paper is the first systematic description of the DBC database machine, and for this reason the first subject the authors deal with concerns design problems. In their opinion the problems met in system design are intrinsically related to the nature of conventional hardware and can only be solved by introducing new architectural concepts. The DBC's functional characteristics and theory of operation are then illustrated and the paper concludes with a high-level description of DBC organization.

    Google Scholar 

  17. J. Banerjee, D. K. Hsiao and K. Kannan: DBC — A Database Computer for Very Large Databases, IEEE Trans. on Computer, Vol. C-28, N. 6, June 1979, pp. 414–429. In this paper, the overall architecture of the DBC machine is described and the organization of the individual components, as well as the implementation of some important concepts which are vital to database management, are discussed. The choice of technologies to be used for implementing the various components of the machine in terms of cost and performance is examined. The paper shows that the DBC machine provides a very high-level instruction repertoire for interfacing with the front-end, a set of elaborate security mechanisms, and an effective cluster mechanism.

    Google Scholar 

  18. P. Bertolazzi, M.Missikoff and M.Terranova: CID: A VLSI Device for List Intersection, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 188–204. A VLSI device for intersecting two unsorted lists with a high degree of parallelism is described and analyzed. This device is designed for intersecting two lists of tuple identifiers representing intermediate results of a query in the DBMAC database machine.

    Google Scholar 

  19. H. Boral, D.J. DeWitt: Processor Allocation Strategies for Multiprocessor Data Base Machine, ACM TODS, Vol.6, n. 2, June 1981, pp. 227–254. Four alternative strategies for assigning processors to queries in multiprocessor database machines are described and evaluated: SIMD assignment, packet-level assignment, instruction level and data flow assignment. Evaluation is performed by means of simulation techniques. The queries are subdivided into classes of varying complexity and then their mixes are examined.

    Google Scholar 

  20. H. Boral, D.J. DeWitt: Database Machines: An Idea Whose Time has Passed? A Critique of the Future of Database Machines, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 166–187. This paper describes three classes of database machines (Processor-per-Track architectures, Processor-per-Head designs and Off-the-Disk machines) and discusses the impact that trends in mass storage and processor technology have on these designs. The authors assert that highly parallel database machine architectures are doomed to extinction unless mechanisms for increasing the bandwidth of mass storage devices are found. Three fields of research are suggested: using unmodified disk drives with a customized disk controller, front-ending a number of conventional disks with a very large and very fast RAM memory, and investigating effective index strategies.

    Google Scholar 

  21. H. Boral, D.J. DeWitt: A Methodology for Database System Performance Evaluation, Computer Science Department Technical Report, Univ. of Wisconsin, 1984. A benchmarking-based methodology for evaluating the performance of database management systems and database machines in a multiuser environment is presented. The authors show that only four basic query types are needed for constructing a benchmark capable of evaluating a system's performance under a wide variety of workloads. This report is an extended version of [BOD84b].

    Google Scholar 

  22. H.Boral, D.J.DeWitt: A Methodology for Database System Performance Evaluation, Proceeding of Annual Meeting SIGMOD'84, Boston, Ma., 1984, pp. 176–185. This paper presents a methodology for evaluating the performance of database management systems and database machines in a multiuser environment. The transaction throughput is studied by developing a methodology for multiuser benchmarks. The authors identify three main factors that affect transaction throughput: multiprogramming level, degree of data sharing among simultaneously executed transactions, and transaction mix. They demonstrate that only four basic query types are needed for constructing a benchmark capable of evaluating a system's performance under a wide variety of workloads. Lastly, they present the results obtained by applying the methodology to the Britton-Lee IDM-500 database machine, see [BRI81].

    Google Scholar 

  23. H.Boral, S.Redfield: Database Machine Morfology, Proc. of 11th Int. Conf. on Very Large Data Bases, Stockholm, 1985, pp. 59–71. This paper analyzes and classifies twenty database machines and catalogue them on the basis of seven macro characteristics: type of mission, number of simultaneous missions, overlap type, memory property, processing primitives, location mechanisms, and storage structures. Furthermore, a language which describes the anatomy of DBM architectures in terms of a collection of modules, links between modules, and subsystems grouping modules is proposed. The analysis points out the following drawbacks: little attention is given to the I/O bottleneck usually faced only by the use of brute force parallelism, the designs are usually optimized towards improving the response time of a single request (in most cases relational operations instead of queries) instead of also being throughput oriented.

    Google Scholar 

  24. O.H.Bray, H.A. Freeman: Data Base Computers, Lexington Books, 1979. This book introduces readers to data base management and database computer concepts. CAFS, CASSM, STARAN, RAP, DBC architectures are surveyed, classified and compared to one another. Their classification is based on the number of processors involved in database processing and on the type of processing used (search for data on mass storage devices or in an intermediate storage area). They use the following classes: single processor direct search, multiple processor direct search, multiple processor indirect search, and multiple processor combined search.

    Google Scholar 

  25. Britton-Lee Incorporation: Intelligent Data Base Machine Product Description, Britton-Lee Inc., 1981, Los Gatos, California. The Intelligent Database Machine (IDM) is an integrated hardware/software backend computer designed to provide quality database performance at a moderate cost. It provides a host-independent facility for managing data and the intelligence required for managing user communication is provided by the host's software. IDM architecture is based on a specially designed processor called the Database Accelerator and the tasks a relational DBMS performs while processing queries are microcoded in it. The IDM is not intended for users requiring extremely high transaction rates.

    Google Scholar 

  26. Britton-Lee Incorporation: IDM Software Reference Manual Version 1.6, Britton-Lee Inc., 1984, Los Gatos, California. The IDM's Intelligent Database Language (IDL) is described here. The IDM contains a relational database management system which is a logical outgrowth of the INGRES system. IDL is available for on-line database creation, accessing and modification. Interface software is available for database management by application programs written in major programming languages and executed in host computers (IDM is not a general-purpose computer and does not have compilers of its own). IDM software features are: relational data management, transaction management, security, optimized access path selection, concurrency control, audit logs, crash recovery, dump and load of data, and a random access file system.

    Google Scholar 

  27. J. D. Brownsmith: A Simulation Model of the MICRONET Computer System during JOIN Processing, Annual Simulation Simposium 1981, pp. 1–16. A simulation study of queueing and resource utilization of MICRONET during processing relational join operations is presented. Some of the results obtained are compared with those obtained by a deterministic model.

    Google Scholar 

  28. J.D. Brownsmith, S.Y.W. Su: Performance Analysis of the Equijoin Operation in the MICRONET Computer System, Proc. of the ICC 80, 1980, p.264–268. The performance analysis is carried out by an analytical model. Results related to a number of tuples ranging from 10⋆⋆2 to 10 ⋆⋆8 and to a number of processors ranging from 1 to 1000 are reported.

    Google Scholar 

  29. A.F. Cardenas, F. Alavian, A. Avizienis: Performance of Recovery Architectures in Parallel Associative Database Processors, ACM TODS, vol. 8, n. 3, September 1983, pp. 291–323. Three different types of recovery mechanisms for parallel associative database processors (belonging to the Processor-Per-Track class) are identified. For each architecture both the workload imposed by the recovery mechanisms on the execution of database operations and the workload involved in the recovery actions are analyzed. The three architectures are then compared to one another in terms of the number of extra database revolutions needed.

    Google Scholar 

  30. F. Cesarini, D. De Luca and G.Soda: An Assessment of the Query-Processing Capability of DBMAC, in "Advanced Database Machine Architecture", D.K.Hsiao Ed., Prentice-Hall, 1983, pp. 109–129. Query processing in the DBMAC database machine is analyzed by means of simulation techniques. Due to the multiprocessor architecture and the particular scheme used for storing the data, two main query schemes, based on selection primitives, are introduced to represent the machine workload. A description of the simulation model and some results obtained by its application are given.

    Google Scholar 

  31. F.Cesarini, F.Pippolini: Parallel Evaluation of Relational Operators in a Data Base Machine, Proc. Int. Symp. MIMI82, Paris, 1982, pp. 19–24. The results concerning the time required for answering two sample queries referred to the DBMAC database machine are given. The data are subdivided into a particular structure, called data pool. A basic set of primitives operating on these data is defined and the transformation of a query parse tree into an executive tree made up of appropriate data primitives is described.

    Google Scholar 

  32. J.P. Cheiney, P. Faudemay, R. Michel and J.M. Thevenin: A Reliable Backend Using Multiattribute Clustering and Select-Join Operator, Proc. 12th Int. Conf. on VLDB, Kyoto, 1986, pp. 220–227. This paper presents some multiprocessor algorithms which speed-up both joins and selections. The techniques proposed are based on the linearization of the join time complexity law. This implies distributing the I/O and the processing load among several processors and disks. The solution proposed in this paper is based on a kind of multi-attribute clustering which uses a parallel implementation of digital hashing and a linearly growing directory. This method is implemented in the SABRE Database Machine. Analytical considerations show that it is possible to improve the multiprocessor hashing join algorithms with a ratio of 3 to 5.

    Google Scholar 

  33. R.McCord: Sizing and Data Distribution for a Distributed Data Base Machine, Proc. ACM SIGMOD, Michigan, 1981, pp. 198–204. A Simulation Program for the Analysis of Database Machines and Environments (SPADE) is described here. This program was made for evaluating the MUFFIN database machine proposal, which is intended to support a distributed version of the INGRES relational database system. The results of the experiments are used for analyzing possible MUFFIN configurations and processing tactics.

    Google Scholar 

  34. F.Cesarini, F.Pippolini, G.Soda: A Technique for Analyzing Query Execution in a Multiprocessor Data Base Machines, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 68–90 (also reprinted in this volume as chapter 9). In this paper, a methodology for representing and evaluating the execution of relational queries by a multiprocessor database machine is presented. The methodology is based on the analysis of a structure called query execution graph. A query execution graph is the detailed description of the way the database machine must operate in order to answer a query. A procedure for computing the execution cost of query by examining the query execution graph is proposed and takes the parallel execution of subqueries into account.

    Google Scholar 

  35. D.J.DeWitt, R.Gerber: Multiprocessor Hash-Based Join Algorithms, Proc. of 11th Int. Conf. on Very Large Data Bases, Stockholm, 1985, pp. 151–164. This paper examines multiprocessor hash-join algorithms in a multiprocessor environment where it is possible to identify CPU, communication and I/O bandwidth design parameters. The algorithms analyzed are multiprocessor versions of Simple, Hybrid, and Grace algorithms previously examined by [DKS84] in a single processor environment. A simulation model is constructed and performance is measured by throughput. Utilization of CPU, disk, and network is also illustrated. The results evidence linear increases in throughput with corresponding increases in processor and disk resources.

    Google Scholar 

  36. D.J.DeWitt, P.B. Hawthorn: A Performance Evaluation of Data Base Machine Architectures, Proc. 7th Int. Conf. on VLDB, Cannes, 1981, pp. 199–213. In this paper, analytical models for a conventional database management system and four generic database machine architectures are proposed. The architectures are classified as PPT (Processor-per-Track systems), PPH (Processor-per-Head systems), PPD (Processor-per-Disk systems) and MPC (Multiprocessor-Cache-Systems) and the following three kinds of queries are taken into consideration: selection, join and aggregate function queries. It is demonstrated that no one type of machine is the best one for executing all types of queries. Furthermore, for some classes of queries, certain database machine designs are slower than a DBMS on a conventional computer.

    Google Scholar 

  37. D.J. De Witt: DIRECT — A Multiprocessor Organization for Supporting Relational Database Management Systems, IEEE Trans. on Computer, Vol. C-28, N. 6, June 1979, pp. 395–406. This paper presents the overall architecture of the DIRECT machine. The author emphasizes the MIMD aspect of the architecture; i.e., the machine can simultaneously support both intra-query and inter-query concurrency. This feature is obtained by means of an associative memory and an interconnection matrix which permits two query processors to search the same page of the same relation simultaneously while executing different queries. Furthermore, the author proposes a dynamic mechanism for determining the number of processors to be allocated to a query. This mechanism is based on the priority of the query, the size of the relations it references, and the type and number of relational operations included in the query. The relation size is not limited by the size of the associative memory.

    Google Scholar 

  38. D.J. DeWitt, R.H. Gerber, G. Graefe, M.L. Heytens, K.B. Kumar and M. Murailikrisna: GAMMA — A High Performance Dataflow Database Machine, Proc. of 12th Int. Conf. on VLDB, Kyoto, 1986, pp. 228–237. GAMMA is a relational database machine that exploits dataflow query processing techniques. Its architecture consists of 20 VAX 11/750 processors connected together by an 80 megabit/second token ring. GAMMA is different from a distributed database system running on a local network because it has no notion of site autonomy, and possesses a centralized schema and a single point for starting the execution of all queries. A preliminary performance analysis based on a benchmark strategy elaborated by [BDT83] is also reported in this paper.

    Google Scholar 

  39. S.A. Demurjian, D.K. Hsiao, D.S. Kerr, J. Menon, P.R. Strawser, R.C. Tekampe, J. Trimble, R.J. Watson: Performance Evaluation of a Database System in a Multiple Backend Configuration, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 91–111. This paper deals with the measurement of a database system running on a multiple backend configuration obtained by means of benchmarking experiments. Benchmarking strategies are devised and applied to the external and internal measurements of a MBDS prototype (see [HHK83]). The external performance measurement of MBDS was obtained by collecting macroscopic data, such as the response time of a request, while the internal performance measurement of MBDS was obtained by collecting microscopic data, such as the entering and leaving time in a system process.

    Google Scholar 

  40. D.J.DeWitt, R.H.Katz, F.Olken, L.D.Shapiro: Implementation Techniques for Main Memory Database Systems, Proc. of Annual Meeting SIGMOD'84, Boston, Ma., 1984, pp. 1–8. In this paper, the authors deal with the changes that must be made in a relational database system so that it can take advantage of large amounts of main memory. In particular, they compare alternative access methods, such as AVL and B+-trees, to one another when applied to main memory database systems, and measure the performance of algorithms used in relational database operations in this environment. Four algorithms for executing joins are presented and evaluated. They are called: Sort-merge, Simple, GRACE and Hybrid algorithm. The multiprocessor versions of these algorithms are discussed in [DEG85].

    Google Scholar 

  41. M. Drawin, H. Schweppe: A Performance Study on Host-Backend Communication, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 135–153 (also reprinted in this volume as chapter 6). Host-backend communication is analyzed by means of a simulation model. It is shown that this subsystem has a strong influence on the performance of the overall system.

    Google Scholar 

  42. P. Faudemay, P. Valduriez: Design and Analysis of a Direct Filter Using Parallel Comparators, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 137–152. A hardware filter without compilation is described. It is based on a comparator array and distributed logic for solving boolean connectors; a prefilter, realized by a specialized VLSI component of the filtering processor, is used in order to extend the functionality of the filter. A deterministic analysis of the filter and comparisons with compiled filters are made; the analysis made concerns selections and semijoins.

    Google Scholar 

  43. D.Ferrari: Computer Systems Performance Evaluation, Prentice-Hall, 1978. The conceptual aspects of performance evaluation techniques are described here and the author does not neglect the informative aspects that contribute significantly to providing a comprehensive view of all existing material in both research and all other fields.

    Google Scholar 

  44. S. Fushimi, M. Kitsuregawa, M. Nakayama, H.Tanaka and T. Moto-Oka: Algorithm and Performance Evaluation of Adaptive Multidimensional Clustering Technique, ACM SIGMOD, Austin, Texas, 1985, pp. 308–318. The clustering algorithm outlined in this paper is based on space splitting technique. This method, denoted as GKD-tree (Generalized KD-tree), is an extension of a KD-tree method proposed by the same authors. The algorithm's performance is analyzed by means of a theoretical analysis and compared to the KD-tree method. It is shown that the GKD-tree method can largely reduce the average number of page accesses. Both methods are implemented on the GRACE Database machine [MOF83].

    Google Scholar 

  45. S. Fushimi, M. Kitsuregawa, H.Tanaka and T. Moto-Oka: Multidimensional Clustering Technique for Large Relational Database Machine, Proc. of Int. Conf. on Foundations of Data Organization, Kyoto, 1985, pp. 226–235. This paper is an extension of the [FNK85] paper and focuses on the theoretical aspects of the algorithm when space is limited.

    Google Scholar 

  46. S. Fushimi, M. Kitsuregawa and H. Tanaka: An Overview of the System Software of a Parallel Relational Database Machine GRACE, Proc. of 12th Int. Conf. on VLDB, Kyoto, 1986, pp. 209–219. GRACE [MOF83] is a parallel relational database machine which is primarily used for join-intensive applications. The system software described here emphasizes the execution and control of relational operations. In the data-stream-oriented processing discussed, its execution and control unit is the whole set of tuples referred to by the operation and not just a single data page. The system software is organized into a hierarchy, and the execution of a relational operation and its operand data are encapsulated and controlled in the form of a task. The data stream control protocol between modules in a task makes the tasks autonomous. Several performance evaluations are conducted by means of a simulator and concern intra-and inter-task control layers. The results obtained are expressed from a qualitative point of view.

    Google Scholar 

  47. D.H.Fishman, M.Y.Lai, W.K.Wilkinson: Overview of the Jasmin Database Machine, Proceeding of Annual Meeting SIGMOD'84, Boston, Ma., 1984, pp. 234–239. In this paper the architecture of a multiprocessor database machine called JASMIN is described. This machine can be configured for several applications and implemented by using "off-the-shelf" parts, and it is able to handle distributed databases efficiently. The performance observed in the uniprocessor prototype is compared to that of a Britton-Lee IDM-500 (see [BRI81]) and is measured in terms of response time for ten test queries in a single-user environment. The measurements are obtained by using the UNIX "time" utility.

    Google Scholar 

  48. G.Gardarin, F.Pasquer: Design and Implementation of Sabre — a Deductive and Parallel Database Machine, in "Database Machines — Modern Trends and Applications", NATO ASI Series, Springer-Verlag, 1986, pp. 203–216. The SABRE machine's design and architecture, as well as a preliminary design of the rule management functions to be added to it, are described in this paper.

    Google Scholar 

  49. S.Gamerman, M.Scholl: Hardware versus Software Data Filtering: the VERSO Experience, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 112–136. This work is concerned with the problem of choosing between a hardware and software approach for implementing the filters used in the VERSO database machine. Hardware filters are implemented by dedicated hardware, while software filters consist in writing a code to be run on an "off-the-shelf" microprocessor. Comparison between these two kinds of filters is made by evaluating the response time to a selection/projection query by means of an analytical model.

    Google Scholar 

  50. G.Gardarin, P.Bernadat, N.Temmerman, P.Valduriez, Y.Viemont: SABRE — A Relational Database System for a Multimicroprocessor Machine, in "Advanced Database Machine Architecture", D.K.Hsiao ed., Prentice-Hall, 1983, pp. 19–35. SABRE is a software-oriented machine developed as a portable system on both big computers and multimicroprocessor machines. Its functional architecture is composed of virtual processors mapped on one or more real processors. On-the-fly filtering, multidimensional clustering, and view mechanisms are some of the main characteristics of the project.

    Google Scholar 

  51. R.Gonzales-Rubio, J.Rohmer: From Databases to Artificial Intelligence: a Hardware Point of View, in "Database Machines — Modern Trends and Applications", NATO ASI Series, Springer-Verlag, 1986, pp. 323–340. This paper discusses the chief similarities existing between Artificial Intelligence and Data Base systems: in both fields, the data to be manipulated are represented in symbols, and both domains require content addressing and setoriented processing. The authors examine the use of filters in a deductive environment, capable of processing on-the-fly data coming from a disk. The SCHUSS filter in particular is examined.

    Google Scholar 

  52. S. Gamerman, S. Salza, M.Scholl: A Methodology for Evaluating the Filter Utilization in the DBM VERSO, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 91–105 (also reprinted in this volume as chapter 8). A probabilistic modelling approach for evaluating filter utilization in the VERSO relational database machine is proposed. An analysis is made of the union operation under a simple algorithm in order to illustrate this approach.

    Google Scholar 

  53. P.B. Hawthorn, D.J. DeWitt: Performance Analysis of Alternative Database Machine Architectures, IEEE Trans. on Software Engineering, Vol. SE-8 n. 1, 1982 pp.61–75. In this paper, a comparison is made among RAP, CASSM, DBC, DIRECT, and CAFS database machines and associative disks by using three benchmark retrieval queries according to the INGRES system. As a result, a comparison between the above-mentioned systems and the INGRES system is also obtained. It is shown that data-intensive queries can be performed very efficiently on database machines if the function performed on the data is entirely provided by the database machine. If it isn't, the host processor is too highly utilized and so the database machine hardly improves the system's performance at all. The same queries are later used to predict the performance of the NON-VON parallel machine applied to databases [HSN86].

    Google Scholar 

  54. R.B. Hagmann, D. Ferrari: Performance Analysis of Several Back-End Database Architectures, ACM TODS, vol.11, n.1, March 1986, pp. 1–26. Some ways of offloading some functions of a database system to a back-end computer are studied experimentally in this paper. The INGRES relational system is divided into six subsystems according to the following functions: User Interface, Query Parser, Query Decomposition and Planning, Inner Loop, Access Methods, and File System. Different strategies of assigning these parts to two conventional computers connected together by a local area network are analyzed by benchmarks. Database and queries represent statistical applications.

    Google Scholar 

  55. P.B.Hawthorn, M.Stonebraker: Performance Analysis of a Relational Data Base Management System, Proc. ACM-SIGMOD, Int. Conf. Management of Data, Boston, Ma, 1979, pp. 1–12. In this paper, the authors study the effects extended storage devices, multiple processors and prefetching data blocks have on data management performance, applied to the INGRES System. The following three sets of benchmark queries are taken into account: overhead-intensive, data-intensive, and multirelation queries. The results obtained by running these benchmarks under the INGRES system suggest that back-end data management machines that distribute processing towards the data are not cost-effective if the application supported is mainly overhead-intensive.

    Google Scholar 

  56. P.B.Hawthorn: The Effect of Target Applications on the Design of Database Machines, Proc. ACM-SIGMOD, Michigan, 1981, pp. 188–197. This paper shows how the design of data manipulation processors is application-dependent. Three database machine classes are studied: single data manipulation processor systems, multiple disk-associative data manipulation processor systems, and multiple cache-associated processor systems. The applications are divided into business, bibliographic search, and statistical analysis applications. The performance of each category of database machines applied to each type of application is analyzed. The performance index is an extension of the usual instruction rate for a computer system which includes the concept of data processing rate.

    Google Scholar 

  57. P.B. Hawthorn: Microprocessor Assisted Tuple Access, Decompression and Assembly for Statistical Database Systems, Proc. 8th Int. Conf. VLDB, Mexico City, Sept. 1982, pp. 223–233. This paper presents a design for a Microprocessor Assist System (MAS), a back-end system that performs part of the work of a statistical database management system. Some of its features such as attribute partitioning, compression and data access, are analyzed, evaluated and compared with conventional systems.

    Google Scholar 

  58. D.K.He, M.Higashida, D.S.Kerr, A.Orooji, Z.Shi, P.R.Strawser, D.K.Hsiao: The Implementation of a Multibackend Database System (MDBS), in "Advanced Database Machine Architecture", D.K.Hsiao ed., Prentice-Hall, 1983, pp. 300–385. This paper describes the general architecture and implementation features of MDBS. The MDBS' architecture divides the database system's work among several backends, each of which executes the same system software. A minicomputer is used as the controller and other minicomputers with disks are used as backends. This can be considered a multibackend software approach because no special hardware is required for it.

    Google Scholar 

  59. S.Hikita, S.Kawakami, H.Haniuda: Database Machine FREND, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 190–207. In this paper, the authors describe a relational database machine called FREND, developed for composing distributed systems by means of personal computers and a local area network. Particular attention is given to its processing structure and to optimization of query execution. FREND's performance is evaluated by measuring its response time with an internal hardware timer. This timer only measures FREND's query execution time and therefore does not take communication or personal computer overhead into account.

    Google Scholar 

  60. Y.C.Hong: A Pipeline and Parallel Architecture for Supporting Database Management Systems, Proc. of Int. Conf. on Data Engineering, Los Angeles, 1984, pp. 152–159. This paper describes a pipeline and some parallel architecture used for supporting efficient execution of projection and join operations. Performance is studied by means of a hardware simulator implemented on a PDP11-70 computer.

    Google Scholar 

  61. D.K.Hsiao ed.: Advanced Database Machine Architecture, Prentice-Hall, 1983. This book examines the following nine database machines: DBMAC (Italy), DSDC (Japan), IQC (Japan), MDBS (U.S.A.), RDBM (Germany), SABRE (France), VERSO (France), a join hardware device for DBC (U.S.A.), and a Full-Text Information-Retrieval System (U.S.A.).

    Google Scholar 

  62. B.K. Hillyer, D.E. Shaw, A. Nigam: NON-VON's Performance on Certain Database Benchmarks, IEEE Trans. on Software Engineering, vol. SE-12, n.4, 1986, pp. 577–583. This paper deals with the performance of a parallel machine called NON-VON when used in database management applications. The analysis follows the indications proposed in [HAD82]: the same database and queries are taken into account, and the execution time on a hardware configuration of NON-VON (comparable to the database machine there examined) is calculated. The analysis shows that NON-VON can reach higher performances, expecially as far as queries involving join operations are concerned.

    Google Scholar 

  63. Intel Corporation: FAST-3805 Semiconductor Disk, Intel Corp., 1982, Austin, Texas. The SYSTEM 2000-FAST 3805 Data Base Assist Processor is intended to be used with a host computer as an intelligent I/O and controller processor. The FAST-3805 is a semiconductor disk memory emulating standard large IBM disks, except that it provides faster access to data and faster transfer rates. It uses a MOS solid-state technology, it requires no electromechanical movement and it provides higher reliability. The database throughput rate and response time are improved by several orders of magnitude.

    Google Scholar 

  64. Proc. of Int. Workshop on Database Machines, Florence, 1981. These papers include the proceedings of the 1st Int. Workshop on Database Machines referred to DBMAC, the Italian project for a database machine. Other presentations concerning with VERSO, RDBM, SABRE, DBC and MICRONET architectures are summarized in the foreword.

    Google Scholar 

  65. Proc. of Int. Workshop on Database Machines, S.Diego, 1982. Revised versions of the papers included in the proceedings of the 2nd Int. Workshop on Database Machines are also published in [HSI83].

    Google Scholar 

  66. H.O.Leilich, M.Missikoff ed.: Database Machines, Springer-Verlag, 1983. This book contains the papers presented at the 3rd Int. Workshop on Database Machines held in Munich in 1983.

    Google Scholar 

  67. D.J.DeWitt, H.Boral ed.: Database Machines, Springer-Verlag, 1985. This book contains the papers presented in the 4th Int. Workshop on Database Machines held in Grand-Bahama Island in 1985.

    Google Scholar 

  68. W. Kim, D. Gajski, D.J. Kuck: A Parallel Pipelined Relational Query Processor, ACM TODS, vol. 9, n. 2, June 1984, pp. 214–242. This paper presents the design of a VLSI relational query processor which consists of four processing PIPEs and some random access memory modules. Each PIPE processes tuples of relations in parallel during the evaluation of relational algebraic operators; PIPEs are functionally specialized. A complex relational query consists of a certain number of components the authors call primitive database operations. Each PIPE is capable of processing a primitive database operation in a pipelined manner. Algorithms for supporting both relational operators and arithmetic and aggregation functions are described.

    Google Scholar 

  69. T. Kakuta, N. Miyazaki, S. Shibayama, H. Yokota and K. Murakami: The Design and Implementation of Relational Database Machine Delta, Proc. of the Fourth Int. Workshop on Database Machine, Grand Bahama Island, Springer-Verlag, 1985, pp. 13–35. This paper deals with the overall architecture, functions, processing algorithms and implementation of DELTA. This machine uses specialized hardware to perform its internal set oriented operations while its architecture design is based on functional decomposition into three main kinds of units: RDBE, a Hierarchical Memory (HM) and a Control Processor (CP). RDBE is a Relational Database Engine capable of processing a tuple [SIK84]. Some RDBE units may be present and each of them may interact with the HM and CP. As far as the connection to the outside world is concerned, DELTA architecture includes a front-end processor (Interface Processor) and one or more units, called Maintainance Processors (MP), are included for system supervising.

    Google Scholar 

  70. M.Y.Lai, W.K.Wilkinson: Distributed Transaction Management in JASMIN, Proceedings of 10th Int. Conf. on VLDB, Singapore, 1984, pp. 466–470. In this paper, the authors discuss distributing data and metadata in JASMIN (see [FLW84]) and describe both the distributed multiversion validation technique and the two-phase commit protocol, which is used for achieving concurrency control and crash recovery for data and metadata.

    Google Scholar 

  71. M.D.P.Leland, W.D.Roome: The Silicon Database Machine, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 169–189. This paper describes the design of the hardware and software for a multiprocessor, silicon memory, database machine called SiDBM. The entire database resides in stable silicon memory and it has no disks. Its processors are functionally specialized, as relation managers, host interfaces, query managers and query processors and are tightly coupled. Its silicon memory is shared and is directly addressable by all of them. This paper also contains some preliminary performance results obtained by means of some benchmark tests proposed in [BDT83a].

    Google Scholar 

  72. C.S. Lin, D.C.P. Smith, J.M. Smith: The Design of a Rotating Associative Memory for Relational Database Applications, ACM TODS, vol. 1, n. 1, March 1976, pp. 53–65. This paper deals with the RARES database machine. Its main feature is that of performing tuple selection operations at the storage device and providing a mechanism for efficient sorting. RARES search logic is attached to the heads of a rotating head-per-track storage device. RARES is different from other designs for rotating associative stores in that it utilizes a novel orthogonal storage layout. It provides a high output rate of selected tuples and an order of magnitude reduction in the capacity of local storage to search logic with respect to other cellular designs.

    Google Scholar 

  73. S.E.Madnick: INFLOPEX-Hierarchical Decomposition of a Large Information Management System Using a Microprocessor Complex, AFIPS Conf. Proc., 1975 NCC, vol. 44. The main characteristic of the INFLOPEX is to use the same functional decomposition which can be accomplished on a large information system. The resulting system can be implemented with low-cost LSI devices. A hierarchical memory structure similar to traditional virtual systems is also described.

    Google Scholar 

  74. J. Madelaine: Performance Evaluation of Concurrency Control Algorithms in the SABRE Database Machine, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 277–292. This paper compares the performance of a two-phase locking concurrency control algorithm to that of a time-stamp ordering one. This comparison is made by solving analytically a queueing network which gives the response times of the SABRE database machine.

    Google Scholar 

  75. S.E. Madnick, Y.R. Wang: Modeling the INFOPLEX Database Computer: a Multiprocessor System with Unbalanced Flows, Proc. of 6-th Advanced Database Symposium, Tokyo, 1986, pp. 85–92. A performance analysis methodology using generalized queueing network models to evaluate the speed performance of INFOPLEX is presented in this paper. This methodology focuses on multiprocessor computer systems in which the number of transactions leaving a server is not the same as the number of transactions entering the server, due to asynchronously spawned parallel tasks. A cost effective software tool is developed according to this methodology for analyzing the architectural design alternatives of INFOPLEX. The authors show that this software tool produces the same quality of results as simulation but with less effort and at a fraction of its time and cost.

    Google Scholar 

  76. M.J.Menon, D.K.Hsiao: Design and Analysis of a Relational Join Operation for VLSI, Proc. 7th Int. Conf. on VLDB, Cannes, 1981, pp. 44–55. A hardware organization which performs relational equality joins in database machine environments is proposed. Queueing analysis of the join operation is also used for obtaining closed-form equations for various design parameters.

    Google Scholar 

  77. M.J.Menon, D.K.Hsiao: Design and Analysis of Join Operations of Database Machines, in "Advanced Database Machine Architecture", D.K.Hsiao ed., Prentice-Hall, 1983, pp. 203–255. This paper proposes an extendable organization of processors and memories for hardware realization of relational join operations. It analyzes the results regarding some equality joins described by [MEH81] in greater detail and covers natural join, inequality join and m-way join. This hardware organization is proposed for the DBC computer and is compared with hardware joins proposed for other database machines.

    Google Scholar 

  78. J. Menon: A Study of Sort Algorithms for Multiprocessor Database Machines, Proc. of 12th Int. Conf. on VLDB, Kyoto, 1986, pp. 197–206. This paper presents and analyzes algorithms for parallel execution of sort operations in a general multiprocessor architecture. These algorithms pertain to both internal and external sorting. As far as the internal sorting algorithm is concerned, the author presents an analysis of a bitonic merge as an alternative to the two-way merge [BDH84]. As far as the external sorting algorithms are concerned, two techniques for improving their performance are suggested: the use of pipelining and the use of parallel internal sorting. A deterministic analysis is performed on three algorithms: the pipelined odd-even sort, the block bitonic sort, and the modified block bitonic sort. The author shows that this last sort is the fastest algorithm for a wide range of values.

    Google Scholar 

  79. M.Missikoff, M.Terranova: The Architecture of a Relational Database Computer known as DBMAC, in "Advanced Database Machine Architecture", D.K.Hsiao Ed., Prentice-halll, 1983, pp. 87–108. The major features of the DBMAC database machine are presented. Its multiprocessor physical architecture is based on general-purpose processing units which can communicate via a global memory. Each processing unit can access to each of the disks via a Mass Memory Bus that provides several parallel colloquies between a processing unit and a disk. Its logical architecture is subdivided into two basic sections called High System and Low System. The first performance evaluation of the High System is also given.

    Google Scholar 

  80. T. Moto-Oka, K. Fuchi: The Architectures in the Fifth Generation Computers, Proc. of he IFIP 9th World Computer Congress, Paris, North-Holland, 1983, pp. 589–602. The database machine called GRACE is presented as a typical example of the knowledge-base machine architecture proposed in the FGCS project. GRACE is a relational algebra machine which adopts a relational algebra processing algorithm based on hash and sort. It can join two relations in O((M+N)/n) time, where n is the number of processors, and M and N are the cardinalities of the two joined relations. The global architecture consists of the following four kinds of modules: processing, memory, disk, and control. These modules are connected to each other by two ring buses.

    Google Scholar 

  81. M.Missikoff, S.Salza, M.Terranova: DBMAC — A Parallel Relational Database Machine, in "Database Machines — Modern Trends and Applications", NATO ASI Series, Springer-Verlag, 1986, pp. 85–126. This paper describes the software and hardware architecture of DBMAC and a specialized device [BMT83] for list intersections. A short description of problems and techniques to be used for evaluating DBMAC performance is also given.

    Google Scholar 

  82. T.Nakayama, M.Hirakawa, T.Ichikawa: Architecture and Algorithm for Parallel Execution of a Join Operation, Proc. of Int. Conf. on Data Engineering, 1984, pp. 160–166. The paper presents an algorithm for the parallel execution of a join operation in a type of architecture composed of one master unit and several slave units linked to each other. The architecture looks similar to the architecture proposed in [MEH81] but its relations are distributed to the slave units by hash. Performance is analyzed in a deterministic way and the static execution cost is then derived.

    Google Scholar 

  83. E.A. Ozkarahan, S.A. Schuster and K.C. Sevcik: Performance Evaluation of a Relational Associative Processor, ACM TODS Vol. 2, n. 2, June 1977, pp. 175–195. A comparative performance evaluation between the RAP database machine and a conventional database management system is presented. Deterministic models are developed for each system. Basic relational DBMS operations, including simple retrieval, updating, computation functions and implicit join, are taken into account in the comparison.

    Google Scholar 

  84. E.Ozkarahan: Database Machines and Database Management, Prentice-Hall, 1986. This book introduces some systems suitable for nonnumeric processing: parallel and pipeline architectures, associative memories, and associative processors. A large number of database machines are surveyed and the main hardware and software issues of the various database architectures are analysed.

    Google Scholar 

  85. E.A.Ozkarahan, M.Ouksel: Dynamic and Order Preserving Data Partitioning for Database Machines, Proc of 11th Int. Conf. on Very Large Data Bases, Stockholm, 1985, pp. 358–368. The authors support the theory that the I/O bottleneck problem cannot be solved by designing additional architecture and then therefore propose a multidimensional data partitioning structure that can enhance cellular and instream architectures. Their method, otherwise called The Interpolation Based Grid File and used for direct addressed files, is adapted for database machines. The use of this partitioning scheme for implementing join and projection in the RAP.3 machine is described.

    Google Scholar 

  86. E.A. Ozkarahan, K.C. Sevcik: Analysis of Architectural Features for Enhancing the Performance of a Database Machine, ACM TODS, Vol. 2, n. 4, December 1977, pp. 297–316. This paper deals with the detailed design and analysis of some mechanisms which improve the RAP's performance. These mechanisms can produce features similar to multiprogramming and virtual memory, which are found in general-purpose computer systems. Expressions for comparing scheduling disciplines are derived. A brief summary is given of experiments concerning virtual memory facility; they are made by means of a simulation model, also illustrated in [SOS76].

    Google Scholar 

  87. G.Z.Qadah: The Equi-Join Operation on a Multiprocessor Database Machine: Algorithms and the Evaluation of their Performance, Proc. of 4th Int. Workshop on Database Machines, Grand Bahama Island, Springer-Verlag, 1985, pp. 35–67. Parallel algorithms for implementing the equi-join operation on the Michigan Relational Database Machine (MIRDM) are presented. A study of the performance of the algorithms proposed is outlined. A probabilistic average-value framework for modeling both the algorithms proposed and MIRDM hardware organization is used in order to determine the overall best performing equi-join algorithm and to investigate the effectiveness of performing some tuning on MIRDM's architecture.

    Google Scholar 

  88. G.Z.Qadah, K.B.Irani: A Database Machine for Very Large Relational Databases, Proceeding of the Int. Conf. on Parallel Processing, 1983, pp. 307–314. The organization of the Michigan Relational Database Machine (MIRDM) is described. It consists of four main components, namely, the master back-end controller (MBC), the processing cluster subsystem (PCS), the mass storage subsystem (MSS) and the interconnection network subsystem (INS). The MBC acts as an interface to the host computer, as a monitor to query execution and as a manager of MIRDM's various components. The MSS is organized as a two level memory. The PCS is a set of SIMD processing clusters. The INS consists of a set of bidirectional buses and connects the other components together.

    Google Scholar 

  89. C.Riechmann: IDM 500 Within A Mainframe Environment — Some First Experiences, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 227–232. This paper briefly outlines some experiences concerning the connection between the IDM-500 machine and a SIEMENS 7561 mainframe under a BS2000 operating system.

    Google Scholar 

  90. G.Schumacher: GEI's Experience with Britton-Lee's IDM, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 233–241. Some measurements on a IDM-500 machine connected with a SIEMENS 7536 mainframe running under a BS2000 operating system are illustrated. These measurements concern data retrieval, creation of indexes and other features offered by IDM software. Some remarks about its usability are also made.

    Google Scholar 

  91. K.C.Sevcik: Data Base System Performance Prediction using an Analytical Model, Proc. 7th Int. Conf. on VLDB, Cannes, 1981, pp. 182–198. This paper suggests an overall framework for predicting and assessing the effect on resource consumption, throughput, and response time of a variety of physical and logical database design decisions that affect performance. At its lowest level, the analytical model is based on queueing networks. At higher levels, a description of a sequence of database system workloads is proposed. The workload description at one level and a set of design choices are transformed into the workload description at the next lower level by means of some anlytical techniques.

    Google Scholar 

  92. R.K. Shultz, R.J. Zingg: Response Time Analysis of Multiprocessor Computers for Database Support, ACM TODS, Vol. 9, n.1, March 1984, pp. 100–132. A comparison is made of three multiprocessor computer architectures (DIRECT, HYPERTREE and REPT) for database support. The algorithms performed by each machine in order to execute a single query involving selection, projection and join operations are analyzed. Deterministic expressions for response time are established and evaluated.

    Google Scholar 

  93. H. Sakai, K. Iwata, S. Kamiya, M. Abe, A. Tanaka, S. Shibayama and K. Murakami: Design and Implementation of the Relational Database Engine, Proc. of Int. Conf. on Fifth Generation Computer Systems 1984, Tokyo, 1984, pp. 419–435. In this paper, the authors describe the Relational Database Engine (RDBE), the key component for processing relational database operations in the database machine DELTA. DELTA's overall architecture is presented in [KMS85]. The basic idea of RDBE processing is that a join operation is efficiently performed by sorting tuples of each relation according to their values and by then comparing the tuples from the sorted relations in a two-way merge way. As a result the RBDE architecture is made up of some sort and merge cells which process a tuple in a pipeline way.

    Google Scholar 

  94. H.Sakai, K.Iwata, S.Shibayama, M.Abe, H.Itoh: Development of Delta as a First Step to a Knowledge Base Machine, in "Database Machines — Modern Trends and Applications", NATO ASI Series, Springer-Verlag, 1986, pp. 159–182. The paper presents a relational database machine called DELTA, that is being developed at the Institute for New Generation Computer Technology. It is based on functionally distributed architecture, relational database engines for performing relational database operations, and a high capacity hierarchical memory system. Expected performance of join and sort operations is analyzed by deterministic formulae. As far as the developing of a knowledge base machine is concerned, DELTA can be regarded as its tabular knowledge component.

    Google Scholar 

  95. S.Y.W.Su, S.Lupkiewics, C.Lee, D.H.Lo and K.L.Doty: MICRONET a Microcomputer Network System for Managing Distributed Relational Databases, Proc. 4th Int. Conf. on VLDB, Berlin, 1978, pp. 288–298. This paper deals with the hardware and software design of the microcomputer network called MICRONET. A preliminary analytical evaluation is made and is compared to conventional systems.

    Google Scholar 

  96. J.Slonim, L.J.McRae, N.Diamond, W.E.Mennie: NDX-100: An Electronic Filing Machine for the Office of the Future, Computer, vol. 14, n. 1, pp. 24–36. The NDX-100 is a prototype system whose architecture is made up of few or many microprocessors operating parallel to each other and concurrently, on a common data storage area in conventional random access devices. The NDX-100 handles the inverted organization for a file; when a query is to be processed, the NDX-100 assigns a set of microprocessors to it made available from a pool of microprocessors used for servicing queries. According to the level of query complexity, different numbers of microprocessors are assigned to perform in parallel, where possible, the subtasks required for answering the query.

    Google Scholar 

  97. S.Y.W. Su, L.H. Nguyen, A. Emam, G.J. Lipovski: The Architectural Features and Implementation Techniques of the Multicell CASSM, IEEE Trans. on Computers, vol. C-28, n.6, 1979, pp. 430–445. This paper describes the context-addressed segment sequential memory system called CASSM. CASSM has a cellular type of architecture in which each cell contains both storage and processing elements. This system also offers associative and parallel processing capabilities for efficient data retrieval and manipulation in large databases. The hardware is designed mainly to support a hierarchical data model but it can also be used in other contexts. The authors give a detailed description of the hardware implementation techniques used in this system.

    Google Scholar 

  98. A.K.Sood and A.H.Qureshi eds.: "Database Machines — Modern Trends and Applications", NATO ASI Series, Springer Verlag, 1986. This book is a collection of 28 papers presented at the NATO Advanced Study Institute School held in Les Arcs, France, in July 1985. The authors are researchers coming from France, Germany, Italy, Japan, Portugal, Turkey, U.K. and U.S.A.

    Google Scholar 

  99. S.A. Schuster, E.A. Ozkarahan and K.C. Smith: A Virtual Memory System for a Relational Associative Processor, Proc. AFIPS NCC, vol.45, 1976, pp.855–862. In this paper, the authors present a virtual memory environment for the RAP database machine and describe its performance results obtained by simulation.

    Google Scholar 

  100. S.A. Schuster, H.B. Nguyen, E.A. Ozkarahan, K.C. Smith: RAP.2 — An Associative Processor for Databases and Its Applications, IEEE Trans. on Computer, Vol. C-28, N.6, June 1979, pp. 446–458. The RAP machine is a multiprocessor back-end whose architecture is based on the fact that database operations are inherently set-oriented and data addressing is best accomplished through associative reference to achieve high data independence. The basic architecture of a RAP device consists of a set of identical cells, a static arithmetical unit and a central controller. The general strategy for parallel processing is SIMD because each RAP instruction is simultaneously executed within the cells which operate directly on the data in parallel. This paper describes the RAP.2 version which is faster and more flexible than the previous one.

    Google Scholar 

  101. L.J. Siegel, H. J. Siegel and P.H. Swain: Performance Measures for Evaluating Algorithms for SIMD Machines, IEEE Trans. on Software Engineering, Vol. SE-8, n. 4, July 1982, pp. 319–331. A number of performance measures for evaluating SIMD algorithms are examined. Although the example given of a SIMD algorithm only concerns the image-processing problem domain, the authors make some very useful remarks on measurement concepts including execution time, parallel efficiency, speed, overhead ratio, processor utilization, redundancy, cost effectiveness, speed-up of the parallel algorithm over the corresponding serial algorithm, and an additional measure called "price", which assigns a weighted value to computations and processors.

    Google Scholar 

  102. G.Schiffner, P.Scheuermann, S.Seehusen and H.Weber: On a Specification and Performance Evaluation Model for Multicomputer Database Machines, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 46–73. In this paper, the authors propose a database machine performance evaluation model made up of the following three interrelated models: a database model, a cost estimation model, and a simulation model based on Petri nets. A technique is given for describing the mapping of a database system to physical processors and the mapping of logical execution schedules to physical execution schedules.

    Google Scholar 

  103. G.Stiege: RDBM — Software Considerations and Performance Evaluation, in "Database Machines — Modern Trends and Applications", NATO ASI Series, Springer-Verlag, 1986, pp. 15–44. The RDBM machine's software is described and an analytical model which uses queueing networks according to a two level modeling technique is briefly outlined.

    Google Scholar 

  104. A.Sekino, K.Takeuchi, T.Makino, K.Hakozaki, T.Doi, T.Goto: Design Considerations for an Information Query Computer, in "Advanced Database Machine Architecture", D.K.Hsiao ed., Prentice-Hall, 1983, pp. 130–167. This paper describes the design and some implementation details of a database machine (IQC) developed at the NEC corporation. Host-IQC interface, architectural choices, requirements for reliability, and integrity and security are the main topics discussed. This machine is intended for use in a distributed processing environment.

    Google Scholar 

  105. S.Salza, M.Terranova and P.Velardi: Performance Modelling of the DBMAC Architecture, Proc. of 3rd Int. Workshop on Database Machines, Munich, Springer-Verlag, 1983, pp. 74–90 (also reprinted in this volume as chapter 7). A performance analysis of the DBMAC database machine is made and a two level hierarchical model for performing internal and global analysis is developed. The lower level model is based on a queueing network which makes it possible to compare different design alternatives while the higher level is more schematic and it provides global indices and workload investigation.

    Google Scholar 

  106. S.Y.W. Su: Cellular-Logic Devices: Concepts and Applications, Computer, vol. 12, n. 3, 1979, pp. 11–25. This paper points out the limitations of conventional disks and describes the general characteristics of cellular-logic devices, and some existing devices and their applications in data retrieval and character string processing. The author discusses the limitations, issues and problems related to these devices. The discussed devices include TapeDRUM, RAPID and CASSM.

    Google Scholar 

  107. S.Y.W. Su, A. Emam: Casdal: CASSM's DAta Language, ACM TODS, vol.3, n.1, 1978, pp. 57–91. In this paper the authors describe the high level Casdal language designed and implemented for database machine CASSM. Its language refers to a unnormalized (hierarchically structured) relational data model and contains constructs both for processing data and processing relations. The constructs are directly supported by the hardware.

    Google Scholar 

  108. S.Y.W. Su, K.P. Mikkilineni: Parallel Algorithms and their Implementation in MICRONET, Proc. 8th Int. Conf. on VLDB, Mexico City, 1982, pp. 310–324. The design and implementation of hardware and software and the parallel algorithms for four categories of database operations are described and illustrated in this paper. Three new algorithms are proposed together with their implementations in MICRONET: one for finding maximum/minimum, and two for sorting distributed files. They are compared with other sorting algorithms.

    Google Scholar 

  109. H. Schweppe, H.Ch.Zeidler, W.Hell, H.O.Leilich. G.Stiege, and W.Teich: RDBM — A Dedicated Multiprocessor System for Database Management., in "Advanced Database Machine Architecture", D.K.Hsiao Ed., Prentice-Hall, 1983, pp. 36–86. The authors describe the architecture of RDBM, a centrally controlled multiprocessor system which contains a content addressable memory and specialized processors for sorting and interrecord operations supporting binary relational operations. The special function processors have common access to a large main memory. The different hardware components are controlled by a general purpose minicomputer which also performs query analysis and optimization.

    Google Scholar 

  110. Y.Tanaka: A Data-stream Database Machine with Large Capacity, in "Advanced Database Machine Architecture", D.K.Hsiao ed., Prentice-Hall, 1983, pp. 168–202. This paper describes the design of the DSDC relational database machine which is based on data encoding, segmentation of files, data-stream processing of files, and data-flow control of segment flows. It contains functionally specialized devices: a binary trie engine for indexes management, a search engine, and a sort engine.

    Google Scholar 

  111. W.Teich: RDBM — Special Hardware Design for Sorting, in "Database Machines — Modern Trends and Applications, NATO ASI Series, Springer-Verlag, 1986, pp. 45–68. This paper describes a hardware sort which is part of RDBM. The sort is based on a 4-way merge technique. The sort device also supports an external sorting facility by merging presorted data blocks.

    Google Scholar 

  112. M.Ubell: The Intelligent Database Machine (IDM), in "Query Processing in Database Systems", W.Kim, D.S.Reiner and D.S.Batory Eds., Springer-Verlag, 1985, pp. 237–247. In this paper the author describes the software and hardware architecture of the IDM machine and discusses the communication between the host computer and IDM and the issues involved in dividing the total workload between the host computer and the IDM backend machine. The author also illustrates some performance characteristics of the IDM and its throughput in a multiuser environment. The tests cited concern retrieving data from 1 to 16 relations, each having 10,000 records.

    Google Scholar 

  113. P. Valduriez, G. Gardarin: Multiprocessor Join Algorithms of Relations, Proc. of 2nd Int. Conf. on Improving Data Base Usability and Responsiveness, Jerusalem, 1982, pp. 219–236. Three algorithms for computing joins in a multiprocessor database machine model based on SABRE architecture are proposed and analyzed. They are based on the nested loop join algorithm, sort merge join algorithm and hashing join algorithm. Performance analysis is obtained by means of deterministic models.

    Google Scholar 

  114. P. Valduriez, G. Gardarin: Join and Semijoin Algorithms for a Multiprocessor Database Machine, ACM TODS, Vol. 9, n. 1, March 1984, pp. 133–161. Algorithms for computing joins and semijoins of relations in a multiprocessor database machine model based on SABRE architecture are proposed. In addition to proposing the analysis of a previous paper [VAG82], a comparison is made between the methods of joining two relations by means of the nested loop join algorithm and by means of semijoins and it is shown that the semijoin method is generally better.

    Google Scholar 

  115. V.Vemuri, R.A. Liuzzi, J. P.Cavano and P.B. Berra: Evaluation of Alternate Database Machine Designs, Proc. 5th Workshop on Computer Architecture for Non Numeric Processing, Pacific Grove, 1980, pp. 29–38. The authors propose a first systematic approach to developing methodologies for performing DBM analysis. The following four classes of evaluation criteria are proposed: performance, cost, quality and human engineering. Each of these criteria refers to three levels of detail: the user level, the system level and the device level. A logical sequence of stages for conducting performance analysis of DBM designs is indicated and includes mathematical modelling, simulation, emulation and statistical analysis of hardware/software measures.

    Google Scholar 

  116. H.Ch.Zeidler: RDBM — A Relational Data Base Machine Based on a Dedicated Multiprocessor System, in "Database Machines — Modern Trends and Applications, NATO ASI Series, Springer-Verlag, 1986, pp. 15–44. The RDBM machine, implemented at the Technical University of Braunschweig (Germany), is thoroughly examined. Special emphasis is given to hardware-supported functions.

    Google Scholar 

Download references

Authors
  1. F. Cesarini
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. F. Pippolini
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. G. Soda
    View author publications

    You can also search for this author in PubMed Google Scholar

Editor information

Francesca Cesarini Silvio Salza

Rights and permissions

Reprints and permissions

Copyright information

© 1987 Springer-Verlag Berlin Heidelberg

About this chapter

Cite this chapter

Cesarini, F., Pippolini, F., Soda, G. (1987). Chapter 10 Database machine design and performance evaluation: Annotated bibliography. In: Cesarini, F., Salza, S. (eds) Database Machine Performance: Modeling Methodologies and Evaluation Strategies. Lecture Notes in Computer Science, vol 257. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-17942-9_20

Download citation

  • DOI: https://doi.org/10.1007/3-540-17942-9_20

  • Published:

  • Publisher Name: Springer, Berlin, Heidelberg

  • Print ISBN: 978-3-540-17942-9

  • Online ISBN: 978-3-540-47140-0

  • eBook Packages: Springer Book Archive

Publish with us

Policies and ethics

Search

Navigation