leff@smu.UUCP (Laurence Leff) (12/14/89)
-----------------------------------------------------------------------------
89-CSE-1. ($1.00)
MODELING TRANSACTIONS IN OBJECT ORIENTED DATA BASES*
Francisco Mariategui and Margaret H. Eich
January 1989
We present a new model for transactions in Object Oriented Data Bases
(OODB). This model captures the shape of transactions by following the
behavior of the message passing mechanism inherent in the object oriented
paradigm. We call this new kind of transaction, "The Pair <GM, DM>". This
dynamic model gives rise to tree structured transactions. The end product of
this structure are data base accesses. Our model builds a bridge between
message passing, by which objects communicate, and data base accesses, by
which transactions communicate. Thereby closing the gap between the dynamic
view of the world (objects) and the static view of the world (data bases).
* This material is based in part upon work supported by the Texas Advanced
Research Program (Advanced Technology Program) under Grant No. 2265 and by
the National Science Foundation under Grant IRI-8713654.
-----------------------------------------------------------------------------
89-CSE-2. ($1.25)
DAA - A PROTOTYPE OF AN INTEGRATED PROTOTYPING SYSTEM (IPS)
Programmer's Reference Manual
W. P. Yin and M. M. Tanik
This document describes details of the implementation of the Design
Activity Agent (DAA). Topics covered are: the project objectives, a
description of built-in maintenance aids in the program, facility
descriptions, a description of Design Object Descriptive Attribute Notation
(DODAN), a list of unimplemented features, a discussion of the rapid
prototyping environment being used, and possibilities for future
modification. The actual source code for the program is included separately.
The program of DAA is written in ART (The Automatic Reasoning Tool) a rule
like language, using forward chaining mechanism. Some portions are written
in Common-lisp. The DODAN description program is written in ART. A large
number of ART utilities are used in the program. It is necessary to read the
ART manuals, to fully understand the program.
-----------------------------------------------------------------------------
89-CSE-3. ($3.50)
DAA - A PROTOTYPE OF AN INTEGRATED PROTOTYPING SYSTEM (IPS)
User's Manual
W. P. Yin and M. M. Tanik
This system implements a prototype of a software design paradigm on top
of the ART. The system is primarily meant to be used as a software design
environment for aiding a software designer to design software programs, but
can also be used as an intelligent assistance to understand software designs.
DAA allows a user to construct a software design in DODAN by using ZMACS
editor, save the design in a file, review and analyze the design.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-4. ($1.00)
THE AT&T VIDEOTEX COMMUNICATION SYSTEM-III AT SOUTHERN METHODIST UNIVERSITY
Bryan Beeman and Murat M. Tanik
February 1989
Project Goal: To bring the AT&T Videotex Communication System at SMU to an
operational state.~--~This report discusses the varied aspects of the system,
beginning with a description of the system, containing both physical and
functional descriptions of each system module. The next part of the report
is a comprehensive systems analysis covering all aspects of the VCS,
concentrating on the major conceptual elements: the frame creation module,
the host computer, and the array of public access terminals. The discussion
then covers the progress of the project toward the project goal and problems
encountered along the way. Following this is a summary of what to expect
from the available documentation.
-----------------------------------------------------------------------------
89-CSE-5. ($1.00)
ACTIVE DESIGN COMPONENTS FOR MULTIPROCESSING APPLICATIONS
M.G. Christiansen and M.M. Tanik
February 1989
~~~~~This paper describes our efforts in creating support for design
applications which require the presentation and manipulation of rapidly
evolving systems, with an emphasis on developing software design
environments. This support takes the form of active design components that
interact and evolve as a community towards an implementation of the
application being developed. An objective of this work is the development of
domain specific abstractions for specific classes of programming problems.
Multiprocessing, or parallel programming provides the basis for the set of
components we have implemented and utilize in the implementation in our
prototype environment.
-----------------------------------------------------------------------------
89-CSE-6. ($2.00)
SMU/SEAS NETWORK SIMULATION IN PAWS/GPSM
Greg S. Moore and Murat M. Tanik
February 1989
~~~~~This project concerns the use of the PAWS software package to evaluate
and generate a performance analysis of the School of Engineering and Applied
Sciences computer network at SMU. The project has been developed over two
semesters because of the large amount of computer network information needed
and the complexity of PAWS. This report outlines the entire procedure
followed during the course of this project. All work has been directed
towards the development of the PAWS (Performance Analysts Workbench System)
program which generates performance analyses of the SEAS computer network.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-7. ($3.00)
REQUIREMENTS SPECIFICATION FOR A REAL-TIME EMBEDDED EXPERT SYSTEM
Sang C. Suh, Frank Coyle, Dennis J. Frailey, Murat M. Tanik
February 1989
~~~~~Several commercial expert system shells, among them Texas Instruments'
PC Plus, provide knowledge engineers with the capability of developing expert
system applications. While such systems provide a range of development
tools, they are not able to meet the size constraints or provide the run-time
performance needed to address problems associated with delivery of embedded
real-time applications. Furthermore, there is no provision to provide
deliverable code in Ada--a requirement for many DoD systems.
~~~~~The Embedded Consultant project is intended to address these issues by a
two-fold approach: a) knowledge base size reduction by means of "code
optimization" techniques; b) an inference engine, written in Ada, designed
for real-time applications. The project was begun at Texas Instruments under
IDEA program funding. The project was transferred to SMU in the Summer of
1988.
-----------------------------------------------------------------------------
89-CSE-8. ($1.50)
WORKING-SET COPROCESSOR*
Milan Milenkovic
March 1989
~~~~~Although the working-set theory provides a sound basis for efficient
management of virtual memory, few if any faithful practical implementations
of it have been reported. The primary obstacle lies in the apparent O(n)
time complexity of per-reference accounting of page usage required by the
definition of the working set, where n is the working-set size of the process
under consideration.
~~~~~This paper describes the algorithmic and architectural design of a
specialized coprocessor capable of tracking the working set of a running
process with 100% accuracy in real time. A customized combination of
hashing, timer processing, and hardware assists is employed to achieve O(1)
average time complexity of the operations involved in per-reference page
accounting.
~~~~~Results of a trace-driven simulation of the working-set coprocessor
(WSC)'s operation are presented to demonstrate the feasibility of the
working-set coprocessor's implementation using components no faster than
those of the cache memory, and to provide some related implementation
suggestions.
*~This work was partially supported by IBM Corporation. Limited distribution
~~of this paper will take place in the form of a technical report published
~~by IBM.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-9. ($1.00)
REQUIREMENTS STUDY FOR A COMPUTING ENVIRONMENT
FOR ENGINEERING EDUCATION USING X-WINDOWS
Frank P. Coyle, Sang C. Suh, Murat M. Tanik
March 1989
~~~~~Preliminary requirements for a network-based computing facility for
engineering education were defined based on interviews with administrative
staff of the School of Engineering. Prototype graphic interfaces based on
Suntools and X-Windows were developed that permitted icon-based access to
system functions. X-Windows was found to be extremely useful in the
development of graphical interfaces and provided insights into possible
extensions of this work in the area of intelligent user interfaces.
-----------------------------------------------------------------------------
89-CSE-10. ($1.75)
UNIT OF CONSISTENCY IN OBJECT ORIENTED DATA BASES*
Francisco Mariategui and Margaret H. Eich
March 1989
~~~~~We present a new model for transactions in Object Oriented Data Bases
(OODB). This model captures the shape of transactions by following the
behavior of the message passing mechanism inherent in the object oriented
paradigm. We call this new kind of transaction, "The Pair <GM, DM>". This
dynamic model gives rise to tree structured transactions. The end product of
this structure are data base accesses. Our model builds a bridge between
message passing, by which objects communicate, and data base accesses, by
which transactions communicate. Thereby closing the gap between the dynamic
view of the world (objects) and the static view of the world (data bases).
*~This material is based in part upon work supported by the Texas Advanced
~~Research Program (Advanced Technology Program) under Grant No. 2265 and by
~~the National Science Foundation under Grant IRI-8713654.
-----------------------------------------------------------------------------
89-CSE-11. ($2.25)
MULTI-GROUP MULTI-LAYER CONCURRENCY CONTROL IN OODBs*
Francisco Mariategui and Margaret H. Eich
March 1989
~~~~~We propose a multi-group multi-layer approach to concurrency control in
Object Oriented Data Bases (OODB). Our proposed architecture provides
flexibility and adaptability in the sense that it allows several concurrency
control techniques to coexist in the same system. OODBs have the potential
to handle a variety of environments and this capability must be matched with
an appropriate mechanism to handle concurrency.
~~~~~We specify the workings of a Concurrency Control Manager Module that
handles several groups of transactions in parallel. We also decompose the
concurrency control mechanism into pieces. Each one of these pieces
accomplish a specific function and represent different levels of abstraction
in concurrency control processing. The resulting module provides the
underlying technology to make our approach feasible.
*~This material is based in part upon work supported by the Texas Advanced
~~Research Program (Advanced Technology Program) under Grant No. 2265 and by
~~the National Science Foundation under Grant IRI-8713654.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-13. ($1.50)
SYSTEM SPECIFICATION WITH COMMUNICATING SEQUENTIAL PROCESSES (CSP)
Sanjive Agarwala Murat M. Tanik
Texas Instruements Inc. Southern Methodist University
April 1989
~~~~~Communicating Sequential Processes (CSP) is an abstract, formal model of
information flow, particularly suited in describing and analyzing systems and
applications that may exhibit multiple concurrent and communicating
activities. This study is aimed at applying the basic ideas of CSP in
hardware design and finding if this approach yields some practical benefits
in terms of coping with complex system design problems, with comparative ease
of design. CSP can provide formalisms that truly express concurrency from a
very high level of behavioral abstraction, where the architect is still in
control. The basic CSP primitives are very suggestive of their possible
realization in terms of hardware components. Moreover, we believe that there
is a commonality in hardware and software design and thus we may find a
significant applicability of the new ideas of structured programming and
CSP to yield some practical benefits for system design and specification.
-----------------------------------------------------------------------------
89-CSE-14. ($2.25)
STUDIES ON MULTI-PEG TOWERS OF HANOI
Sam S. Chen, Ching-chao Liao*, David Y.Y. Yun, Murat M. Tanik
April 1989
~~~~~A heuristic approach named bipartite consecutive-disk decomposition is
proposed as a divide-and-conquer problem solving paradigm so that its optimal
can be the upper bound solution on the required number of moves for the
multi-peg Towers of Hanoi, i.e. the problem of moving n disks from a source
peg to a destination peg with the aid of k additional working pegs. After
limited exploring on some interesting cases of small n, two heuristic
evaluation functions in terms of certain u, the degree of working peg
utilization, are presented for the optimal decomposition and the effective
prediction of the required number of moves. The predication behavior in
terms of the heuristic evaluation functions is analyzed further. The
required number of moves is shown to be of degenerate exponential with
respect to the number of disks while using more working pegs (k>1)
successively. If we can let k grow at a certain constant rate of n1/u/u,
then the solution behavior can even be linear with 2un.
*~Visiting scholar at SMU in 1988, Professor, Chung Cheng Institute of
~~Technology, Taiwan, Republic of China.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-15. ($2.00)
AN AGENT FOR INTELLIGENT MODEL MANAGEMENT
John I.C. Liu and David Y.Y. Yun
Dept of Computer Science and Engineering
Gary Klein
Dept of Management Information Sciences
Edwin L. Cox School of Business
August 1987 (Revised April 1989)
~~~~~Decision Support Systems (DSS) and Expert Systems (ES) are both aimed at
improving decision-making. Current DSS usually concentrate on quantitative
model methods while ES emphasize logic and reasoning. In this paper, we
review previous approaches used by DSS to handle decision models. We propose
an extension to such systems, an Agent for Intelligent Model Management
(AIMM), as an interface between DSS and the users. AIMM utilizes a
combination of knowledge representations and reasoning mechanisms to make a
wide variety of models available to decision makers so that they may apply
these models without becoming involved in technical or procedural aspects of
implementation. A prototype of the system for financial problems is
described.
-----------------------------------------------------------------------------
89-CSE-16. ($1.75)
LCF: A LEXICOGRAPHIC BINARY REPRESENTATION OF THE RATIONALS
Peter Kornerup David W. Matula*
Odense University Southern Methodist University
April 1989
~~~~~A new binary representation of the rationals derived from their
continued fraction expansions is described and analysed. The concepts
"adjacency", "mediant" and "convergent" (best rational approximation) from
the literature on Farey fractions and continued fractions are suitably
extended to provide a foundation for this new binary representation system.
Worst case representation-induced precision loss for any real number by a
fixed length representable number of the system is shown to be at most 19% of
bit word length, with no precision loss whatsoever induced in the
representation of any reasonably sized rational number. The representation
is proposed as host for a computer arithmetic synergistically supporting
exact rational and approximate real computation.
*~This research was partially supported by the National Science Foundation
~~under grant DCR-8315289.
-----------------------------------------------------------------------------
89-CSE-17. ($1.00)
EXECUTE LOCKING*
Margaret H. Eich
~~~~~An extension to two-phase locking which synchronizes the execution of
procedures is proposed. This technique, Execute Locking, is two-phase with
respect to data access but is not two-phase with respect to procedure
execution. It is shown that serializability is maintained.
*~This material is based in part upon work supported by the Texas Advanced
~~Research Program (Advanced Technology Program) under Grant No. 2265 and by
~~the National Science Foundation under Grant IRI-8713654.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-18. ($1.50)
RAPID PROTOTYPING SYSTEM SPECIFICATION
Joseph S. Dancses and Murat M. Tanik
May 1989
~~~~~We present System Level Specification of a Rapid Prototyping System
including development environment, a programming language, and the human
interface in accordance with the guidelines provided by DoD-STD-2167A.
-----------------------------------------------------------------------------
89-CSE-19. ($1.00)
AN EXPOSE-AND-MERGE ALGORITHM AND THE CHROMATIC NUMBER OF A RANDOM GRAPH*
David W. Matula and Ludek Kucera**
May 1989
~~~~~A randomized algorithm is described employing the expose-and-merge
paradigm to create and color random graphs. An analysis of the algorithm is
then employed to resolve a conjecture about the chromatic number of a random
graph, where specially we prove that almost surely
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
~*~This paper is to appear in the Proceedings from the Third Seminar on
~~~Random Graphs and Probabilistic Methods in Combinatorics and Computer
~~~Science held in Poznan, Poland Aug. 1987 (to be published by Wiley).
**~Professor Kucera is on leave from Charles University, Prague, Czechosolvakia.
-----------------------------------------------------------------------------
89-CSE-21. ($1.50)
REAL-TIME EMBEDDED EXPERT SYSTEMS AND KNOWLEDGE-BASE TRANSLATION PROBLEM
Dennis J. Frailey, Sang C. Suh, Frank Coyle, Murat M. Tanik
June 1989
~~~~~A wealth of tools are now available for expert system development. Most
of these utilize powerful computing resources, including fast processors,
symbolic programming languages, large main memories, and extensive
interactive graphics support. Although well suited to expert system
development, such computing systems are inappropriate for certain embedded
real-time environments in which limited memory sizes, conventional processor
architectures, standard programming languages and serious reliability
considerations apply. Moreover, embedded systems typically have little or no
need for interaction with an operator, but have important requirements for
time-critical, interactive operation with other programs and equipment.
~~~~~This report focuses on automatic translation of knowledge bases into
structures that can be manipulated in conventional programming environments.
Several opportunities are described for accomplishing this goal. Techniques
from compiler code optimization are identified that can facilitate
translation of knowledge bases into compact data structures suitable for
processing with a small inference engine. An experimental effort using an
Ada inference engine is described. Potential problem areas of real-time
systems are classified in an appendix.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-22. ($3.50)
THE LISP PROGRAMMING LANGUAGE: TUTORIAL AND EXPERIENCE
Terry L. Colvin, Frank Coyle, Murat M. Tanik
June 1989
~~~~~LISP was developed in the late 1950's by John McCarthy, making it one of
the oldest programming languages still in use. It was designed specifically
for list processing--the manipulation of symbolic information. Because of
LISP's power and flexibility, it is the most widely used language in the
field of artificial intelligence.
~~~~~Although there are several dialects of LISP, no single one has been
accepted as a uniform standard. The versions of LISP used in this paper will
be Common Lisp, defined by Guy Steele in his book Common Lisp: The Language
and Franz LISP, which was developed in the early 1980's at the University of
California at Berkeley.
-----------------------------------------------------------------------------
89-CSE-23. ($2.00)
THE EVOLUTION OF RANDOM GRAPHS
Michal Karonski*
June 1989
~~~~~Let us adopt a dynamic view in which a graph, acquiring edges, grows
from empty to complete. More precisely, start with n labelled vertices
{1, 2, ..., n} and add edges at random, one at a time, until the resulting
graph is complete. This simple model of a graph evolving in a random manner
was first investigated in detail by Erdos and Renyi (1960). The key question
they posed may be stated in general terms as follows: what is a typical
structure of a random graph during its evolution? We say that a structure
(or a graphical property) is typical if a random graph posseses it almost
surely, i.e., with probability tending to one as the number of vertices tends
to infinity. This whole section is, to a large extent, devoted to the
description of various problems arising from this basic question.
*~Professor Karonski is on leave from the Department of Discrete Mathematics,
~~Adam Mickiewicz University, Poznan, Poland.
-----------------------------------------------------------------------------
89-CSE-24. ($1.00)
MICROPROCESSOR MEMORY-MANAGEMENT UNITS
Milan Milenkovic
June 1989
~~~~~This report describes the current crop of commercial memory management
units (MMUs) for 32-bit microprocessors of both CISC and RISC varieties. The
purpose of this study is to review and compare the design principles and
features of high-end MMUs across a common set of criteria. In addition to
covering traditional support for virtual memory, consideration is given to
support for Unix-style of memory management and for multiprocessor/multiple
MMU configurations.
~~~~~Section 1 of this report contains an overview of the rationale,
principles, and issues related to hardware support for memory management and
virtual memory. Section 2 contains brief overviews of the following MMUs:
Intel~80386, 486, 860, Motorola MC68851 (MMU for MC68020), MC68030,
MC68040, MC88200 (MMU for 88K series), SPARC MMU Fujitsu MB86920, and MIPS
R2000/R3000. Discussion and closing remarks are presented in Section 3 which
also contains a tabulated summary of the presented MMUs.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-25. ($1.00)
ON HAMMING NETWORKS
Wen-Kuang Chou and David Y.Y. Yun
June 1989
~~~~~A revised Hamming net is proposed, validated, and tested. Both the
original Hamming net and the revised model have been implemented.
Experimental results show the equivalence of both models in the noiseless
cases, but show dramatic improvements in classification problems with noise.
A proof of the convergence theorem for the revised model is given.
-----------------------------------------------------------------------------
89-CSE-26. ($1.25)
INTERFACE DESIGN ISSUES IN A LABORATORY SETTING
Mary Lynne Morrow, Maria Fafouti-Milenkovic, Pari Sadagousheh, Murat M. Tanik
June 1989
~~~~~This is a report of the research conducted on behalf of Abbott Hospital
Supply. It states the particular "needs" of Abbott that the research is
addressing and the specific objectives and goals for the first phase of the
research project. The report includes data from a pilot study (interview
responses of lab technicians from different hospital and laboratory sites)
and the preliminary analysis of the results collected so far.
-----------------------------------------------------------------------------
89-CSE-27. ($2.50)
A SYSTEMS DESIGN APPROACH TO A CONCURRENT EMBEDDED
SYSTEM DESIGN PROCESS AND ENVIRONMENT
Geoffrey C. Hingle and Murat M. Tanik
June 1989
~~~~~This paper presents a "System Design-system" (SD-system) which
efficiently integrates the people involved in the management, design,
development, testing, production and maintenance processes of large-scale
embedded systems. The design of large embedded hardware/software systems
such as an aircraft requires the involvement of many people before design
requirement decisions can be made. System requirements must be developed and
negotiated between people and corporations concurrently as the embedded
systems of a large system are developed. Therefore, design methods such as
structured design can be used more efficiently in the system design of these
large embedded systems, if the structured design methodologies support the
design interaction of many people and corporations. The people involved in
embedded system design are the 'glue' that connect the design processes of
each embedded system development into a single process for the design of a
total system of embedded systems. We believe, in order to instantiate
efficient system engineering methods during the concurrent development of
large embedded systems, we must design a SD-System which efficiently connects
the "Human-Objects" (system designers) involved in the concurrent design,
development, test, production, and maintenance of a large embedded system.
This is the main thrust of our paper.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-28. ($1.00)
A SIMULATION TECHNIQUE TO CONSTRUCT PARALLEL EXECUTION PROFILE
FROM SEQUENTIAL TRACE OF LOGIC PROGRAMS*
Chin-Feng Fan and Prasenjit Biswas
April 1989
~~~~~A fast way of estimating the performance of different parallel execution
models for logic programs is presented. This technique constructs parallel
execution profile from sequential trace information. A performance tool can
be conveniently realized with which the speed up, effect of varying
granularity of tasks, parallel execution overheads, and distribution of tasks
can be examined without actually executing benchmark programs. This approach
includes three steps: acquiring sequential trace of a Prolog program,
converting it into task trace, and executing the program based on the
simulated model using the trace. The difficulties of this process are caused
by the backtracking and nondeterministic features of logic programs.
Conceptually a multi-dimensional execution tree of a benchmark program is
reconstructed from its sequential trace for a simulated model. This
technique saves a substantial amount of evaluation time in comparison to the
frequently used technique of simulation/emulation based on instruction-by-
instruction execution of complied logic programs. It can be realized as a
general simulation tool to gain further time saving. Visualization of the
resulting execution tree can easily be included in the future. The technique
has been implemented to compare two different parallel models supporting
different granularities of tasks.
*~This research was supported by Texas Advanced Technology Program Contract
~~3128 (1988).
Appeared in Proc. of 22nd Annual Simulation Conf., Pittsburgh, May l989.
-----------------------------------------------------------------------------
89-CSE-29. ($1.00)
ABSTRACT MACHINE LORAP II AND EXPERIMENTS IN PROCESS GRAIN SIZE DETERMINATION
FOR PARALLEL EXECUTION OF LOGIC PROGRAMS*
Chin-Feng Fan and Prasenjit Biswas
May 1989
~~~~~In this paper we propose a distributed multiprocessor execution model
LORAP II for parallel execution of logic programs. Each processor in the
abstract machine is capable of supporting processes of variable grain sizes
and is designed to be competitive with existing sequential implementations
for deterministic cases. We present a set of experiements that establish the
requirement for determining appropriate process grain sizes for effective
parallel processing. Some heuristics are presented for compile time grain
size determination. Finally, we present some results from our experiements
using the heuristics, that indicate a significant improvement in performance
over a similar model (LORAP) supporting fine grain processes.
*~This research was supported by Texas Advanced Technology Program Contract
~~3128 (1988).
A shorter version has appeared in Proc. of IEEE Symp. on Tools for AI
(Architecture, Languages and Systems), Virginia, Oct. 1989.
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
89-CSE-30. ($1.00)
A MODEL FOR MULTIPROCESSOR I/O
Milan Milenkovic
July 1989
~~~~~Recent advances in processor architecture and memory technology will
likely continue to contribute to increases of computational power of computer
systems. However, the lack of corresponding improvements in input/output
throughput results in somewhat unbalanced systems designs.
~~~~~This paper proposes an architectural approach for increasing the
performance of I/O, especially in multiprocessors and in distributed systems.
The proposal calls for a considerable increase of I/O subsystem autonomy and
self-sufficiency by means of encapsulation of I/O hardware and software into
dedicated processing complexes termed server engines. The server-engine
approach is suited to the specific needs of input/output, including
longevity, burstiness, event-driven and time-critical nature. The paper
describes the basic structure and several different types of server engines.
~~~~~The functional decomposition and encapsulation of I/O operations within
server engines provide some significant benefits in multiprocessor systems.
The major ones include improved I/O performance, increased parallelism, and
significant reduction of the load on the shared system-interconnection paths.
-----------------------------------------------------------------------------
89-CSE-31. ($2.50)
DATABASE PARTITIONING TECHNIQUES TO SUPPORT RELOAD IN A
MAIN MEMORY DATABASE SYSTEM: MARS
Le Gruenwald and Margaret H. Eich
September 1989
~~~~~In a main memory database system, the primary copy of the database is
placed in volatile memory. When a crash occurs, a partial or complete reload
of the database from archive memory (AM) into main memory (MM) is needed. To
effectively perform the reload process without degrading the system
performance, the most effective technique for structuring AM and MM should be
determined. This paper reports on experiments performing a complete analysis
of possible partitioning techniques in terms of the number of I/Os for reload
and number of MM references incurred during transaction processing. Our
analysis shows that horizontal and single vertical partitioning are actually
the only possible candidates. Which one is best depends on the types of
transactions executed.
-----------------------------------------------------------------------------