Analysis of Decision Support for Distributed Systems


written by Francis Chan

In a distributed electronic design system, where tools, components and services are distributed throughout the world, the need for an understanding of the nature of relationships between components and their use in particular designs, related services, and various users or user groups, is essential. The needs can be satisfied by a combination of tools in the area of:

  • Data Collection -- locating useful components and information
  • Data Mining -- capabilities of identifying and classifying various forms of dynamic and static information about the objects and its environment
  • Decision Support -- applications which help the user in decision-making
  • Data Representation -- use and design of data forms for efficient operation and user manipulation
  • Data Visualization -- displaying information in a way of most use to the user

    In this paper, we will briefly explore the concepts and significance of each of the above areas and will look at research and products wherever applicable. Through the investigation of existing applications, we hope to either incorporate these tools into our existing platform and future CAD program and/or learn about the underlying techniques, requirements and essentials capabilities that these tools possess in order to help us design and implement such tools ourselves.


    Data Collection

    Data collection comprises the first stage of system analysis. It starts with problem definition followed by the determination of the kind of relevant data to gather for analysis. The suitable storage medium (e.g. memory, text file, relational database entries, etc.) should also be chosen to allow efficient browsing and manipulation of data. The system designer must also determine the means and locations to getting the required data, such as through textual research, customer tracking data, web site hit sources and patterns, locations of web objects, etc..


    Data Mining

    Data Mining is a process done by a user or a software program in search for patterns of information in a database.
    It may be done reactively -- User Initiative, where the user thinks of a set of specific questions to ask and patterns to look for, and then it's left for the database to perform the relevant queries. Its shortcoming is that it very difficult to think of all the relevant things to look at and very often, important trends are being left unnoticed.
    It can also be done proactively -- System Initiative, where an intelligent software, which may have an underlying engine using neural networks, generates questions by itself, analyzes the database, and looks for patterns without instructions by the user.
    There is also a Mixed-Initiative mode where the user and the system interact, with the system providing some assistance, e.g. Multi-dimensional Analysis with Corporate Vision; the user can look at the data from several viewpoints, with suitable prompts from the system.

    Data mining applications may also perform other activities:

    Predictive Modeling: Patterns discovered from the database are used to predict the future. Predictive modeling allows the user to submit records with one or more unknown values while the system "predicts" the unknowns based on previous patterns of the database.

    Forensic Analysis: Extracted patterns are used to find anomalous, or unusual data elements. To discover the unusual, the application first finds what the norm is, then it detects the items that deviate from the usual outside of a given threshold.

    Data Mining and Intelligent Databases are now becoming a significant market and areas of research. Some people have suggested that the technology is in fact ahead of the application areas and that there is a need for customization of current client base systems to provide solutions within specific industries such as marketing, banking, manufacturing, etc. Moreover, with the maturity of the technologies, data mining over the Internet can be made possible as there is also a growing demand for Internet services in not just providing directory services of where discrete information may be found but more in-depth investigations of trends of users and transactions, which treats the Internet as a large database.

    A Data Mining Application: Information Discovery System IDIS

    Information Discovery, Inc. is an innovative leader and provider of data mining oriented decision support software and solutions, strategic consulting, and warehouse architecture design.

    It main product IDIS: The Information Discovery System has the following characteristics:

  • Uncovers Patterns by Itself
  • Works on Many Computers and Databases
  • Generates Fully Readable English Text
  • Makes Predictions
  • Finds Errors and Anomalies
  • Works on Large Databases and Parallel Computers


    Decision Support

    Decision support systems (DSS) are computer based information systems that are designed with the purpose of improving the processes and outcomes of human decision making. Most decision support systems are developed to support an individual decision maker, such as a manager or an engineer, but some are used by a number of decision makers to arrive at a common decision (termed group decision support systems). An example of a group decision support system is the executive information systems (EIS), which gives many decision makers access to the same information system and supports individual or discrete decisions.

    A Decision Support Application: Runtime Design Automation VOV

    VOV is a design flow manager which automatically captures and manages the dependencies between the files generated during the Computer Aided Design of complex systems, either hardware or software.

    Most design management tools, such as CFI's Encapsulation Standard (TES), emphasizes the control of the design activity by defining one or few templates for each tool, in various forms, like makefile templates, graphs, or rules. Many decisions about design flow and data types as well as project structure have to be made before the design starts. Not only are the preparation of these templates time consuming, it is also difficult, if not impossible, to represent all input/output dependencies of a tool using a template. VOV, however, provides the designer with the flexibility of alternating design methodologies while ensuring that tools are still called in the correct order and that no data are accidentally lost.

    Benefits that VOV presents to the user:

  • Automatic parallel retracing to regain consistency of data after one or multiple changes
  • Documentation of design flow
  • Capability of determining effect of changes before making them
  • Management of computing resources such as computers and floating licenses
  • Archival, retrieval of methodologies, usage and composition of scripts and makefiles
  • Parallel execution of tools on the network
  • Intelligent tools, e.g. clevercopy which avoids re-compiling if there are only changes in comments
  • Command line, menu driven and graphical Tcl/Tk user interfaces
  • Generic interface to databases. Included are UNIX, AR, OCT
  • Support of tools from Cadence, Exemplar Logic, Model Technology, Synopsys, Vantage, Xilinx, UNIX utilities, C and C++ compilers
  • Multiple platform support, AIX, HP-UX, Linux, SunOS-4.x, Solaris-2, HAL


    Block diagram of VOV architecture.


    The client/server architecture of VOV supports concurrent activities, team coordination, distributed data management, and distributed processing.

    Server It manages the design trace. Depending on the operating system, the server can handle up to 250 clients simultaneously. There are four classes of clients:

  • Tools: connect to the server to declare their inputs and outputs, thus allowing the server to build the design trace
  • Users: connect to the server to query and possibly modify the trace.
  • Slaves: offer computing resources to the server
  • Proxies: contribute knowledge about databases

    Dependency graph

    VOV uses a technique called run-time tracing, which is based on the observation that only the tool knows accurately the dependencies implied by its use. VOV offers two ways for tools to communicate the dependency information at run-time:

    Encapsulation: consists normally of a shell script that computes the run-time dependencies, communicates them to the dependency manager, and then calls the tool. It is a powerful technique that does not require any modification of the tool.

    Integration: An option for those who have access to the source code of the tool; it is faster and easier than encapsulation. The VOV Integration Library (VIL) consists of 600 lines of C code distributed in source format.
    (The integration library is passive if VOV is not running; the integrated tool behaves exactly as the original tool.)

    The information produced by the tools is used by the VOV server to build the dependency graph, also called the design trace.


    A design flow graph.


    A design trace which shows the dependencies between 1948 files and 835 tool invocations required to produce the Linux and IBM-AIX version of VOV.



    Data Representation

    Data representation is the abstraction of the concept to be represented. The choice of data structures (at the system level) for representation is very important as it dictates how the system level interacts with the user interface. A good representation should possess the following qualities:

  • Efficiency: the representation (e.g. b+tree, database) should be chosen such that accesses and queries can be done efficiently.
  • Flexibility/Portability: an abstraction should be general enough to allow extensive use in different parts of a model and in different kinds of models.
  • Clarity: a representation should bear a clear and well-defined relationship to the object representing.
  • Transparency: both designers and end-users should be able to understand the model without being aware of the level of abstraction of a particular object.


    Data Visualization

    Computer-Aided Visualization can be divided into two categories: scientific visualization and intuitive data visualization.

    Scientific Visualization strives to convert large data sets into comprehensible pictures. Data sets generated by modern instruments and supercomputers can be so large that visualization is essential to comprehension. The results may require detailed examination by the engineer or scientist of the data as the results might not be obvious to a general audience. Scientific visualization usually involves a great deal of numerical programming: vector arithmetic, inverting matrices, solving ODE's, etc.

    Intuitive Data Visualization is concerned with gaining insight and understanding through visual interpretations of complicated data structures. Its goal is to explain or promote easily understood and comprehended results to a wide audience. Intuitive data visualization is an evolving technology which provides technical innovations that present data in unique and nontraditional ways. Data types, formats and techniques applied include, but is not limited to, multi-dimensional descriptions and animated depictions. Though visualizations are frequently a result of gridded numerical computations, non-geometric calculations and experimental data can also produce animations involving data. While still images and interactive data exploration remain as powerful visualization options, video animation is emerging as a popular way to illustrate concepts.

    Useful properties of a user interface for data visualization include:

  • User Manipulation and Editing
  • Multi-user Support
  • Network Connectivity
  • Interactivity

    Examples of current data visualization applications:

  • Advanced Visual Systems (AVS)

  • IBM Visualization Data Explorer (DX)

  • IRIS Explorer

    Reference Papers:

    On VOV

    [1] Andrea Casotto and Alberto Sangiovanni-Vincentelli. "Automated design management using traces." IEEE Transactions on CAD, August 1993.

    On Data Mining

    Intelligent Database Tools and Applications, New York: John Wiley and Sons, 1993 .

    The Four Spaces of Decision Support, DBMS Magazine, November 1995.

    The Sandwich Paradigm, Database Programming and Design, April 1995.

    What Can IDIS Do That Statistics Cannot?, Information Discovery, Inc., 1995.

    Large Scale Data Mining in Parallel, DBMS Magazine, February 1995.

    Concentric Design for Decision Support, Database Programming & Design, May 1993.

    Information made Visual using HyperData, AI Expert, September, 1992.

    Quality Unbound, Database Programming and Design, November 1994.

    On Intelligent CAD

    Holden, T. The Application of Knowledge-based Systems to Microelectronic CAD (Doctoral Thesis). London University, 1985.

    Holden, T. Knowledge-based CAD and Microelectronics (book). Elsevier North-Holland, 1987.

    Holden, T. Decision Support for Creative Design. Workshop on the Representation of Design in CAD. Centre for Configurational Studies, Open University, Milton Keynes, UK, 7-8 May 1987.

    Holden, T., Flynn, M., Patel, M. A Hardware Synthesis Methodology. Alvey CAD037 project report, Sept 1987.

    Patel, M., Flynn, M., Holden, T. VLSI Design by Intelligent Transformation of Specification. Electronic Design Automation Conference, 1987.

    Brumfit, J., Flynn, M., Patel, M., Holden, T. A Hardware Synthesis Methodology. IEE Design Synthesis Conference, 1988.

    Holden, T. An Intelligent Assistant for the Interactive Design of Complex VLSI Systems. In Yoshikawa, H., Holden, T., Intelligent CAD: Proceedings of the IFIP Second Workshop on Intelligent CAD, Sept 1988. Elsevier North-Holland.

    Holden, T. Overview of Intelligent CAD Issues. Intelligent CAD: Proceedings of the IFIP Second Workshop on Intelligent CAD, Sept 1988. Elsevier North-Holland.

    Yoshikawa, H., Holden, T. Intelligent CAD, Proceedings of the IFIP Workshop on Intelligent CAD, 1990. Elsevier North-Holland.

    Holden, T. Frameworks for Design using Artificial Intelligence Techniques. IFIP TC5/WG5.2 Third International Workshop on CAD, Osaka, Japan, 26-29 Sept 1989.

    Green, I., Holden, T. A Transformational Programming Assistant. In Yoshikawa, H. and Arbab, F., editors, Intelligent CAD III, Procedings of the IFIP TC5/WG5.2 Third International Workshop on CAD, Osaka, Japan, 26-29 Sept 1989, pp129-141. North-Holland, 1991.

    Back to Index


    Modified: February 23, 1995
    Feedback: Francis Chan (fchan@ic.eecs.berkeley.edu)