Manhoi Choy

Efficient fault tolerant algorithms for resource allocation in distributed systems

Solutions to resource allocation problems in distributed systems are examined with respect to the measures of response time, message complexity, and failure locality. Response time measures the time it takes for an algorithm to respond to the requests of a process, message complexity measures the number of messages sent and received by a process, and failure locality characterizes the size of the network that is affected by the failure of a single process. An algorithm that achieves a constant failure locality of four along with a quadratic response time and a quadratic message complexity is presented.

Efficient fault-tolerant algorithms for distributed resource allocation

Solutions to resource allocation problems and other related synchronization problems in distributed systems are examined with respect to the measures of response time, message complexity, and failure locality. Response time measures the time it takes for an algorithm to respond to the requests of a process; message complexity measures the number of messages sent and received by a process; and failure locality characterizes the size of the network that is affected by the failure of a single process. An algorithm for the resource allocation problem that achieves a constant failure locality of four along with a quadratic response time and a quadratic message complexity is presented. Applications of the algorithm to other process synchronization problems in distributed systems are also demonstrated.

Mixed consistency: a model for parallel programming (extended abstract)

A general purpose parallel programming model called mixed consistency is developed for distributed shared memory systems. This model combines two kinds of weak memory consistency conditions: causal memory and pipelined random access memory, and provides four kinds of explicit synchronization operations: read locks, write locks, barriers, and await operations. The resulting suite of memory and synchronization operations can be tailored to solve most programming problems in an efficient manner. Conditions are also developed under which the net effect of programming in this model is the same as programming with sequentially consistent memory. Several examples are included to illustrate the model and the correctness conditions.

Distributed database design for mobile geographical applications

Advanced Traveler Information Systems (ATIS) require efficient information retrieval and updating in a dynamic environment at different geographical scales. ATIS applications are useful in yielding a better utilization of the limited costly transportation arteries and providing value-added traveler information. Many ATIS applications are built on the functionalities provided by Geographical Information Systems (GIS), which often cannot meet extra requirements like real-time response. We investigate GIS-based systems in ATIS and propose a system architecture based on GIS and distributed database technology. Issues on data modeling, data representation, storage and retrieval, data aggregation, and parallel processing of queries are discussed. This paper introduces a distributed system architecture for ATIS based on recent technology. It presents new data models for information representation and proposes data shipping for efficient query processing and function shipping for reducing communication overhead. The paper also examines the use of a network of computers for solving complex problems more timely and privacy protection for sensitive data.

Disaster recovery techniques for database systems

The widespread use of computers has brought about revolutionary changes in society. Computers are becoming vital in all aspects of human life, whether employed in life-critical systems such as air traffic control and autopilot navigation control systems, or in point-of-sales management systems and cinema ticket purchasing systems. Data stored in computer systems is often a company’s most valuable asset, one that must be protected at all costs. Businesses also must be prepared to provide continued service in case of a disaster. Fault-tolerance techniques have been employed to increase computer system availability, and to reduce the damage caused by component failure. Vital data is stored on stable storage, which survives failures such as electrical outages or system crashes. Also, redundant copies of data can be placed on multiple stable storage devices. This approach protects data if failures in storage media are independent, but may be ineffective if disaster strikes. Recall the 1906 earthquake in San Francisco, which destroyed more than half the city. When the U.S. Federal Building in Oklahoma City was bombed, data as well as on-site backups were destroyed. Since data losses and system unavailability resulting from a disaster cripple the operation of an organization, federal legislation now requires the development of recovery plans [5]. Extensive backup procedures have been developed to protect against data losses during disasters, such as the grandfather-father-son backup procedure, the incremental logging technique, and the data image dumping method. In addition to guarding against data losses, a system must also provide its normal services after a disaster strikes. Thus, as with data, computer hardware must also be replicated.

Localizing failures in distributed synchronization

The fault-tolerance of distributed algorithms is investigated in asynchronous message passing systems with undetectable process failures. Two specific synchronization problems are considered, the dining philosophers problem and the binary committee coordination problem. The abstraction of a bounded doorway is introduced as a general mechanism for achieving individual progress and good failure locality. Using it as a building block, optimal fault-tolerant algorithms are constructed for the two problems.

Adaptive solutions to the mutual exclusion problem

Algorithms for mutual exclusion that adapt to the current degree of contention are developed. Afilter and a leader election algorithm form the basic building blocks. The algorithms achieve system response times that are independent of the total number of processes and governed instead by the current degree of contention. The final algorithm achieves a constant amortized system response time.

Maya: a simulation platform for distributed shared memories

Maya is a simulation platform for evaluating the performance of parallel programs on parallel architectures. It allows the rapid prototyping of memory protocols with varying degrees of coherence and facilitates the study of the impact of these protocols on application programs. The design of Maya and its simulation mechanism are discussed. Performance results on architectural simulation with different memory coherence protocols are presented. Parallel discrete event simulation techniques are adopted for the execution-driven simulation of parallel architectures.

Efficient distributed algorithms for dynamic channel assignment

The efficiency of channel assignment in a cellular telephone system is considered using the measures of successful channel assignment ratio as well as response time. Existing paradigms of mutual exclusion and dining philosophers from distributed systems are used to synthesize new algorithms that optimize both measures. The results are verified by extensive simulations.

Robust distributed mutual exclusion

A token based algorithm for distributed mutual exclusion is presented. It uses a distributed counter to tolerate faults due to site failures and communication failures. This eliminates the need for expensive election protocols that are commonly employed in existing token-based algorithms. As in many existing fault-tolerant mutual exclusion algorithms, timeouts are used to detect failures. Our algorithm has the unique property that even if timeout periods are incorrectly assumed, the safety requirement of mutual exclusion is still guaranteed. It is, therefore, suitable for highly asynchronous distributed environments. Performance analysis shows that the algorithm is also efficient in terms of the average number of messages required per critical section, and the average time delay to enter the critical section. Finally, we show how to obtain a bounded algorithm from the basic unbounded algorithm.