Electronic data processing--Distributed processing

System modeling has the potential to enhance system design productivity by providing a
platform for system performance evaluations. This model must be designed at an abstract
level, hiding system details. However, it must represent any subsystem or its components
at any level of specification details. In order to model such a system, we will need to
combine various models-of-computation (MOC). MOC provide a framework to model
various algorithms and activities, while accounting for and exploiting concurrency and
synchronization aspects. Along with supporting various MOC, a modeling environment
should also support a well developed library. In this thesis, we have explored various
modeling environments. MLDesigner (MLD) is one such modeling environment that
supports a well developed library and integrates various MOC. We present an overview
and discuss the process of system modeling with MLD. We further present an abstract
model of a Network-on-Chip in MLD and show latency results for various customizable
parameters for this model.

Parallel/distributed systems offer a tremendous processing capacity. However, in order to take full advantage of it, good load distributions are needed. We study the task graph partition problem for a given parallel/distributed application which is modeled using data parallelism and is implemented in a transputer system in a mesh construction. Our approach uses domain partition to separate a given domain into a set of subdomains of equal size and each of which has at most four neighbors. We devise three methods to partition a given domain, these methods are compared based on several criteria. The impact of the number of processors used in implementation is also investigated based on several parameters, including processor speed, communication speed, and amount of computation and communication per data point. We discuss implementation of our approach in the application based on the existing features of the transputer system, and compare different versions of application through running a simulation system.

In this thesis, a low interprocessor communication overhead and high performance data parallelism parallel application model in a network of workstations (NOWs) is proposed. Checkpointing and rollback technologies are used in this model for performance enhancement purpose. The proposed model is analyzed both theoretically and numerically. The simulation results show that a high performance of the parallel application model is expected. As a case study, the proposed model is used to the parallel Everglades Landscape Fire Model (ELFM) code which was developed by South Florida Water Management District (SFWMD). The parallel programming environment is Message-Passing Interface (MPI). A synchronous checkpointing and rollback mechanism is used to handle the spread of fire which is a dynamic and irregular component of the model. Results show that the performance of the parallel ELFM using MPI is significantly enhanced by the application of checkpointing and rollback.

Current object-oriented development methodologies do not introduce distributed system architectural aspects early enough in the system development cycle. A development partitioning scheme that includes the system level, in addition to the problem and application levels, will encourage analysis that includes elicitation of non-functional system requirements. These requirements include response time with respect to system communication load, fault tolerance, safety, security, and real-time deadlines, among others. They can be documented with an extended form of Jacobson's use cases. Where use cases describe how a system will work from a user's point of view, extended use cases add the capability to describe how well it should work. System level analysis information can be graphically depicted on extended forms of Unified Modeling Language (UML) interaction diagrams and on multilevel architecture diagrams.

Based on multi-agent supporting system (MASS) structures used to investigate the synchronous algorithms in my previous work, the partially and totally asynchronous distributed algorithms are proposed in this thesis. The stability of discrete MASS with asynchronous distributed algorithms is analyzed. The partially asynchronous algorithms proposed for both 1- and 2-dimensional MASS are proven to be convergent, if the vertical disturbances vary sufficiently slower than the convergent time of the system. The adjacent error becomes zero when the system converges. It is also proven that in 1-dimensional MASS using the proposed totally asynchronous algorithm, the maximum of the absolute value of the adjacent error is non-increasing over time. Finally, the simulation results for all the above cases are presented to demonstrate the theoretical findings.

Software for Autonomous Underwater Vehicles (AUVs) is usually complex. It involves several
complicated tasks such as controller, path planner, map builder, and sensor processor.
Distributed processing is unavoidable for such a complex software system with real-time response
requirements. Once these processes are distributed over several computers, it is essential
that they have some mechanism to communicate to each other to share information.
Each process might communicate with several of the other processes. Interprocess communication
becomes an important issue. This thesis discusses the design and implementation
of a generic distributed toolkit that facilitates the development of distributed software for
AUVs. This toolkit makes the details ofinterprocess communication transparent to the programmers
involved in the AUV software development. The toolkit provides efficient direct
communication between processes and does not impose any constraints on the architecture
of the distributed software. Conventional techniques for monitoring/debugging of sequential programs are not applicable
for distributed processes. This is because the correctness of a real-time distributed program
is determined not only by its "logical" correctness but also its "timing" correctness.
Monitoring mainly comprises of observation of the system during its operation. Debugging
is mostly a graphical "post-mortem" analyses of the dumps generated by the distributed program.
This thesis also presents the design of a window based generic graphical monitoring/
debugging tool.

We propose a methodology to effectively characterize the architecture and system performance of distributed systems designed to operate in frame-based real-time environments. Important characteristics that define the real-time performance of a distributed system are identified and classified at the hardware, operating system, and user application levels. A synthetic workload model, called the Distributed Real-Time Workload (DRTW), is designed to fully characterize a broad range of real-time applications and to exercise a single- or multiple-node distributed system under measurement. A set of data collection tools to obtain empirical performance data at different levels of a distributed system is also proposed. For the purpose of illustration, these tools are used to obtain data on several real-time systems from Encore Computer Corporation.

The Course Scheduling Support System is designed to facilitate manual generation of the faculty course scheduling process. It aids in assigning faculty to courses and assigning each course section to their time block. It captures historic and current scheduling information in an organized manner making information needed to create new schedules more readily and quickly available. The interaction between user and database is made as friendly as possible so that managing, manipulating, populating and retrieving scheduling data is simple and efficient. We have implemented an open source web-based prototype of the proposed system using PHP, MySQL, and the Apache Web Server. It can be invoked with a standard Web browser and has an intuitive user interface. It provides tools for customizing web forms that can be easily used by non-technical users. Our department plans to deploy this system by Fall 2006.

Distributed computing has a lot of complex distributed control functions like mutual exclusion, data replication, load sharing, check pointing etc. The present study introduces two novel ways to address some of the issues in the first two areas---mutual exclusion and data replication with the use of mobile agent technology. This research work deals with analyzing the various advantages that Mobile Agent Technology provides and how its use can impact the design and development of distributed system. The primary focus of this study is to present a clear view of the evolutionary path that will take us from the current technology to widespread use of mobile code and agents within the next few years.

Multimedia applications incorporate the use of more than one type of media, i.e., voice, video, data, text and image. With the advances in high-speed communication, the ability to transmit multimedia is becoming widely available. One of the means of transport for multimedia in distributed networks is Broadband Integrated Services Digital Network (B-ISDN). B-ISDN supports the transport of large volumes of data with a low error rate. It also handles the burstiness of multimedia traffic by providing dynamic bandwidth allocation. When multimedia is requested for transport in a distributed network, different Quality of Service (QOS) may be required for each type of media. For example, video can withstand more errors than voice. In order to provide, the most efficient form of transfer, different QOS media are sent using different channels. By using different channels for transport, jitter can impose skews on the temporal relations between the media. Jitter is caused by errors and buffering delays. Since B-ISDN uses Asynchronous Transfer Mode (ATM) as its transfer mode, the jitter that is incurred can be assumed to be bounded if traffic management principles such as admission control and resource reservation are employed. Another network that can assume bounded buffering is the 16 Mbps token-ring LAN when the LAN Server (LS) Ultimedia(TM) software is applied over the OS/2 LAN Server(TM) (using OS/2(TM)). LS Ultimedia(TM) reserves critical resources such as disk, server processor, and network resources for multimedia use. In addition, it also enforces admission control(1). Since jitter is bounded on the networks chosen, buffers can be used to realign the temporal relations in the media. This dissertation presents a solution to this problem by proposing a Feedback-based Multimedia Synchronization Technique (FMST) to correct and compensate for the jitter that is incurred when media are received over high speed communication channels and played back in real time. FMST has been implemented at the session layer for the playback of the streams. A personal computer was used to perform their synchronized playback from a 16 Mbps token-ring and from a simulated B-ISDN network.