Yann Hang Lee

Optimal Checkpointing of Real-Time Tasks

Analytical models for the design and evaluation of checkpointing of real-time tasks are developed. First, the execution of a real-time task is modeled under a common assumption of perfect coverage of on-line detection mechanisms (which is termed a basic model). Then, the model is generalized (to an extended model) to include more realistic cases, i.e., imperfect coverages of on-line detection mechanisms and acceptance tests. Finally, we determine an optimal placement of checkpoints to minimize the mean task execution time while the probability of an unreliable result (or lack of confidence) is kept below a specified level. In the basic model, it is shown that equidistant intercheckpoint intervals are optimal, whereas this is not necessarily true in the extended model. An algorithm for calculating the optimal number of checkpoints and intercheckpoint intervals is presented with some numerical examples for the extended model.

A nonblocking algorithm for shared queues using compare-and-swap

Nonblocking algorithms for concurrent objects guarantee that an object is always accessible, in contrast to blocking algorithms in which a slow or halted process can render part or all of the data structure inaccessible to other processes. A number of algorithms have been proposed for shared FIFO queues, but nonblocking implementations are few and either limit the concurrency or provide inefficient solutions. The authors present a simple and efficient nonblocking shared FIFO queue algorithm with O(n) system latency, no additional memory requirements, and enqueuing and dequeuing times independent of the size of the queue. They use the compare & swap operation as the basic synchronization primitive. They model their algorithm analytically and with a simulation, and compare its performance with that of a blocking FIFO queue. They find that the nonblocking queue has better performance if processors are occasionally slow, but worse performance if some processors are always slower than others. >

Measurement and Application of Fault Latency

The time interval between the occurrence of a fault and the detection of the error caused by the fault is divided by the generation of that error into two parts: fault latency and error latency. Since the moment of error generation is not directly observable, all related works in the literature have dealt with only the sum of fault and error latencies, thereby making the analysis of their separate effects impossible. To remedy this deficiency, we 1) present a new methodology for indirectly measuring fault latency, 2) derive the distribution of fault latency from the methodology, and 3) apply the knowledge of fault latency to the analysis of two important examples.

Error Detection Process—Model, Design, and Its Impact on Computer Performance

Conventionally, reliability analyses either assume that a fault/error is detected immediately as it occurs, or ignore damage caused by imperfect detection mechanisms and error latency, namely, the time interval between the occurrence of an error and the detection of that error.

Design and Evaluation of a Fault-Tolerant Multiprocessor Using Hardware Recovery Blocks

In this paper we consider the design and evaluation of a fault-tolerant multiprocessor with a rollback recovery mechanism.

RTSOA: Real-Time Service-Oriented Architecture

This paper extends the traditional service-oriented architecture (SOA) to a new real-time SOA (RTSOA) and proposes a framework for the new architecture, by providing real-time service modeling, design, code generation, simulation, deployment, execution, orchestration, and management. The concepts, architecture, and enabling technique are studied for real-time SOA. In addition, an efficient algorithm is derived in the paper to find the optimal composition of real-time services subject to the real-time constraint

Voltage-Clock Scaling for Low Energy Consumption in Fixed-Priority Real-Time Systems

Power and energy constraints are becoming increasingly prevalent in real-time embedded systems. Voltage-scaling is a promising technique to reduce energy and power consumption: clock speed tends to decrease linearly with supply voltage while power consumption goes down quadratically. We therefore have a tradeoff between the energy consumption of a task and the speed with which it can be completed. The timing constraints associated with real-time tasks can be used to resolve this tradeoff. In this paper, we present two algorithms for voltage-scaling. Assuming that a processor can operate in one of two modes: high voltage and low voltage, we show how to schedule the voltage settings so that deadlines are met while reducing the total energy consumed. We show that significant reductions can be made in energy consumption.

A study of two-phase service

Some problems in distributed system control, such as load balancing, routing, scheduling in a real-time environment, and reconfiguration require two-phase execution at a central server. That is, jobs come into the server which then probes the distributed system for status information. This is the first phase. The second phase consists of the server performing individual service on each job, e.g., deciding which processor to allocate that job to, or how to reconfigure the system in reaction to the incoming jobs. Here an analysis of such two-phase systems is provided.

Evaluation of Error Recovery Blocks Used for Cooperating Processes

Three alternatives for implementing recovery blocks (RBs) are conceivable for backward error recovery in concurrent processing. These are the asynchronous, synchronous, and the pseudorecovery point implementations. Asynchronous RBs are based on the concept of maximum autonomy in each of concurrent processes. Consequently, establishment of RBs in a process is made independently of others and unbounded rollback propagations become a serious problem. In order to completely avoid unbounded rollback propagations, it is necessary to synchronize the establishment of recovery blocks in all cooperating processes. Process autonomy is sacrificed and processes are forced to wait for commitments from others to establish a recovery line, leading to inefficiency in time utilization. As a compromise between asynchronous and synchronous RBs we propose to insert pseudorecovery points (PRPs) so that unbounded rollback propagations may be avoided while maintaining process autonomy. We developed probabilistic models for analyzing these three methods under standard assumptions in computer performance analysis, i.e., exponential distributions for related random variables. With these models we have estimated 1) the interval between two successive recovery lines for asynchronous RBs, 2) mean loss in computation power for the synchronized method, and 3) additional overhead and rollback distance in case PRPs are used.

Optimal Dynamic Control of Resources in a Distributed System

Abstrad-The various advantages of distributed systems can be reallzed unly when their resources are ‘*optimally” (in some sense) contrdled and utilI7~d. For example, distributed systems must he recnnfigured dynamically to rope with component failures and workload changts. Owing to the inherent dilkulty in formulating and solving re~iurce contrd problems, the resnurce control strategies currently propoacdlused for distributed systems are largely ad hoc. It is our purpose in this paper to I) quantitatively formulate the problem of controlling resources in a distributed system so as tu optimize a reward function, and 2) derive optimal control strategies using Markov decision theory. The control variables treated here are quite generak fnr example, they could be control decisions related to system configuration, repair, diagnostics, files, or data. Two algnrithms for resource contnfl In distributed systems are derived for time-invariant Bnd periodic environments, respectively. A detailed example to demonstrate the power and usefulness of our approach is provided.