Downloads & Free Reading Options - Results

The objective of this project was to contribute to the establishment of the scientific foundation for designing fault-tolerant distributed computer systems. The results obtained from the Phase I of this research project are summarized here. Main results are as follows. (1) A preliminary approach to specification of timing constraints during distributed computer system design and to validation of the temporal behavior of the system. (2) A scheme for coordinated execution of independently designed recoverable distributed processes, and (3) Establishment of initial versions of three of the real time computer network testbeds and a testbed-based evaluation of the distributed recovery block (DRB) scheme. The Phase II of this project was conducted at the University of California, Irvine under Contract No. N00014-87-K-0231 and concluded on Dec. 31, 1989.

Section 1: Introduction -- Introduction -- The basic building blocks -- Section 2: Techniques -- Basic techniques -- Response time -- Time sharing systems -- Reliability and standby -- The man-machine interface -- Conversations with a computer -- The organization of data in the processing unit -- The programs -- The flow of work through a typical system -- Section 3: Applications -- Basic functions of an on-line system -- The real-time computer in a technical system -- The information system in a business environment -- Airline reservations -- Banking -- Message switching -- Manufacturing control -- Section 4: Hardware -- The computers -- Telecommunication facilities -- Terminals and displays -- Planning the data files -- Section 5: Design calculations -- The critical factors -- The estimating process -- Simulation -- Probability and queuing theory -- Estimating the random-access file sizes -- Estimating file reference times and queues -- Estimating the program sizes -- Estimating the system timing -- Estimating the core storage requirements -- Estimating storage for input/output buffering -- Estimating line and terminal requirements -- The reasons for wrong estimates -- Section 6: Implementation -- Accuracy and security -- The programming work -- System testing -- Cutting over the system -- A questionnaire for planning -- Implementation charts

Our research funded under ONR contract N00014-87-K-0796 can be broadly divided into five areas: (1) design and analysis of deadline based scheduling policies, (2) design and evaluation of high performance and fault tolerant disk architectures for real-time systems, (3) design and evaluation of scheduling policies for real-time tasks with incremental-value-with-increased- execution-time characteristics, (4) reliability and testing of real-time systems, (5) scheduling for real-time parallel processing systems. These topics will be the subject of the reminder of the technical section. Additional details of our work can be found in the cited technical papers and reports.

The objective of this project was to contribute to the establishment of the scientific foundation for designing fault-tolerant distributed computer systems. Main results obtained from this research project are as follows. (1) Identification of critical research issues and some promising research directions in real-time fault-tolerant distributed computing; (2) A skeleton of the foundation for realizing system-level fault tolerance, which includes among others the DRB (distributed recovery block) scheme, the DCONV (distributed conversation) scheme, the PTC (programmer-transparent coordination) scheme, a TB (temporary blackout) handling scheme, and the complementary relationship among the schemes; These schemes enable the computer system to detect and recover from both hardware and software faults without missing the deadlines for processing important data and delivering outputs to the controlled object/environment; (3) A preliminary structure of a model of real-time distributed computation; (4) A theoretical investigation into the efficiency and diagnostic power of basic processor-level diagnosis approaches in diagnosing hypercubes conducted; (5) An enhancement of three of the real-time computer network testbeds established in the UCI DREAM (Distributed Real-time Ever Available Microcomputing) Laboratory made.

A new method to determin an optimal retry policy and for use in retry of fault characterization is presented. An optimal retry policy for a given fault characteristic, which determines the maximum allowable retry durations to minimize the total task completion time was derived. The combined fault characterization and retry decision, in which the characteristics of fault are estimated simultaneously with the determination of the optimal retry policy were carried out. Two solution approaches were developed, one based on the point estimation and the other on the Bayes sequential decision. The maximum likelihood estimators are used for the first approach, and the backward induction for testing hypotheses in the second approach. Numerical examples in which all the durations associated with faults have monotone hazard functions, e.g., exponential, Weibull and gamma distributions are presented. These are standard distributions commonly used for modeling analysis and faults.

A new method to determin an optimal retry policy and for use in retry of fault characterization is presented. An optimal retry policy for a given fault characteristic, which determines the maximum allowable retry durations to minimize the total task completion time was derived. The combined fault characterization and retry decision, in which the characteristics of fault are estimated simultaneously with the determination of the optimal retry policy were carried out. Two solution approaches were developed, one based on the point estimation and the other on the Bayes sequential decision. The maximum likelihood estimators are used for the first approach, and the backward induction for testing hypotheses in the second approach. Numerical examples in which all the durations associated with faults have monotone hazard functions, e.g., exponential, Weibull and gamma distributions are presented. These are standard distributions commonly used for modeling analysis and faults.