The concept of atomic transactions has been used to provide reliable processing in both centralized and distributedsystems. An extension of traditional atomic transactions is presented: nested transactions. Nested tr...
详细信息
The concept of atomic transactions has been used to provide reliable processing in both centralized and distributedsystems. An extension of traditional atomic transactions is presented: nested transactions. Nested transactions are seen to permit safe concurrency within as well as among transactions, and to enable transactions to fail partially in a graceful and controlled manner. These properties of nested transactions suit them to a number of distributed applications. Examples of such applications are described.
Task and file allocation are examined in two classes of fault-tolerant distributedsystems. The task allocation problem arises in software-implemented fault tolerance (SIFT)-like systems. while the file allocation pro...
详细信息
Task and file allocation are examined in two classes of fault-tolerant distributedsystems. The task allocation problem arises in software-implemented fault tolerance (SIFT)-like systems. while the file allocation problem arises in Ethernet-like systems. Both problems may be formulated as a constrained sum of squares minimization problem. The computational complexity of these problems prompts us to consider an efficient approximation algorithm that does not always yield optimal answers. It is shown that the ratio of the approximate to the optimal solution is bounded by 9m/8(m minus r plus 1), where m is the number of processors (file servers) to be allocated and r is the number of times each task (file) is to be replicated. Experience with the algorithm suggests that ever better performance ratios can be expected.
An implementation of a reliable remote procedure call (RPC) mechanism for obtaining remote services is described. The reliability issues are discussed together with how they have been dealt with. The performance of th...
详细信息
An implementation of a reliable remote procedure call (RPC) mechanism for obtaining remote services is described. The reliability issues are discussed together with how they have been dealt with. The performance of the remote call mechanism is compared with that of local calls. The remote call mechanism is shown to be an efficient tool for distributed programming.
作者：
Lin, James J.Liu, Ming T.Ohio State Univ
Dep of Computer & Information Science Columbus OH USA Ohio State Univ Dep of Computer & Information Science Columbus OH USA
The system design and performance evaluation of a local data network for very large distributeddatabases. The growing database problem stimulates the need of hardware support for data management in distributed system...
详细信息
The system design and performance evaluation of a local data network for very large distributeddatabases. The growing database problem stimulates the need of hardware support for data management in distributedsystems. A novel hardware configuration, the distributed Double-Loop Data Network (DDLDN), is exemplified. Concurrency control mechanisms and query processing techniques used in the DDLDN are described. Optimal strategy for disk allocation is selected. A performance comparison is made for different types of systems.under various conditions, showing superior performance of the DDLDN. Finally, a way to cope with potential growth of the system is demonstrated.
Some ideas on the construction of user applications as atomic actions are developed. Atomic actions that last a long time pose several problems if conventional ideas on concurrency control and recovery are applied. Wh...
详细信息
Some ideas on the construction of user applications as atomic actions are developed. Atomic actions that last a long time pose several problems if conventional ideas on concurrency control and recovery are applied. What is required is some means of delaying commitment without sacrificing performance. A model is proposed in which it is possible for an action to release and process as yet uncommitable objects. The impact of this on recovery is also discussed.
作者：
Minoura, ToshimiUniv of Southern California
Dep of Electrical Engineering-Systems Los Angeles CA USA Univ of Southern California Dep of Electrical Engineering-Systems Los Angeles CA USA
Two schemes for realizing mutual exclusion in a distributed system are discussed. The ranking scheme is a common framework for the three mutual exclusion algorithms found in the literature. In the ranking scheme, the ...
详细信息
Two schemes for realizing mutual exclusion in a distributed system are discussed. The ranking scheme is a common framework for the three mutual exclusion algorithms found in the literature. In the ranking scheme, the site with the smallest rank number can enter the mutual exclusion state. Another method of realizing mutual exclusion is to let a single 'control token' float in the system and to allow only its holder to enter the mutual exclusion state. One result in this paper shows how the control token is effectively transferred by the ranking scheme.
Intuition tells us that in a distributed DBMS using two phase locking, the ratio (denoted by R/W) of read-only to update transactions affects system performance - the higher the ratio, the better the performance. Read...
详细信息
Intuition tells us that in a distributed DBMS using two phase locking, the ratio (denoted by R/W) of read-only to update transactions affects system performance - the higher the ratio, the better the performance. Read-only transactions only request share locks, and thus should cause fewer conflicts and deadlocks among all transactions. Therefore both read-only and update transactions are expected to perform better if R/W is higher. The results of a study contradicting this intuition are reported, and the relationship between the R/W ratio and system performance in detail.
Two-Phase Commit and other distributed commit protocols provide a method to commit changes while preserving consistency in a distributeddatabase. These protocols can cope with various failures occurring in the system...
详细信息
Two-Phase Commit and other distributed commit protocols provide a method to commit changes while preserving consistency in a distributeddatabase. These protocols can cope with various failures occurring in the system. But in case of failure they do not guarantee termination (of protocol processing) within a given time: sometimes the protocol requires waiting for a failed processor to be returned to operation. It happens that a straightforward use of timeouts in a distributed system is fraught with unexpected peril and does not provide an easy solution to the problem. Byzantine Agreement is combined with Two-Phase Commit, using observations of Lamport to provide a method to cope with failure within a given time bound. An extra benefit of this combination of ideas is that it handles undetected and transient faults as well as the more usual system or processor down faults handled by other distributed commit protocols.
The Byzantine Generals problem involves a system of N processes, t of which may be unreliable. The problem is for the reliable processes to agree on a binary value sent by a 'general' which may itself be one o...
详细信息
The Byzantine Generals problem involves a system of N processes, t of which may be unreliable. The problem is for the reliable processes to agree on a binary value sent by a 'general' which may itself be one of the N processes. If the general sends the same value to each process, then all reliable processes must agree on that value but in any case, they must agree on the same value. An explicit solution is given for a binary value among N equals 3t plus 1 processes, using 2t plus 4 rounds and O(t**3 log t) message bits, where t bounds the number of faulty processes. This solution is easily extended to the general case of N greater than equivalent to 3t plus 1 to give a solution using 2t plus 5 rounds and O(tN plus t**3 log t) message bits.
This paper presents a pragmatic example of how to use Ada tasks with Add machines such as SuperMicro(TM) for scientific computing. First, a class of asynchronous concurrent Newton's algorithms: Purely-Asynchronous...
详细信息
ISBN:
(纸本)9780897910835
This paper presents a pragmatic example of how to use Ada tasks with Add machines such as SuperMicro(TM) for scientific computing. First, a class of asynchronous concurrent Newton's algorithms: Purely-Asynchronous (PAN), Asynchronous (AN), and Semi-Asynchronous (SAN) algorithms are introduced for distributed computing environment. A sufficient condition is shown to guarantee local convergence of these asynchronous algorithms. Moreover, it is proved that the PAN algorithm is linearly convergent, the AN algorithm is super-linearly convergent, and the SAN algorithm is quadratically convergent. The order of convergence depends on the degree of interaction among computing modules. second, a non-deterministic program for the asynchronous algorithm is provided by using Ada tasks. The asynchronous algorithm/program is superior to the synchronized counterpart with respect to computation speed and reliability. Lastly, this paper describes a way where the asynchronous algorithm/program can be realized as a hierarchic distributed system with Ada machines. A preliminary result is shown with MicroAda/SuperMicro(TM).
暂无评论