Studying fo the Efficiency Lmproving Technique of the Highly Available System.

实现高可用性系统是非常重要的，它使用户不知道由软件错误和硬件故障产生的暂时停止。通常，硬件机制用于获得高可用性。通过使用软件技术，获得了以下优点.由软件错误引起的暂时停机可以处理。由于不需要专门的硬件，可以经济地构造高可用系统，但由于同步冗余计算和比较每次计算的结果需要一定的开销，降低了系统的效率。在本研究中，我们发展了一种并发控制机制，借由使用冗余计算来提高高可用系统的效率。以下是我们的结果。1.冗余计算的并发控制。我们通过尽可能异步地执行冗余计算来减少同步冗余计算的开销。我们 ...更多信息 已经证明了通过在后一个系统中按照在快一个系统中执行的相同顺序执行操作序列来减少死锁和回滚，并通过扩展这一思想开发了多等待两阶段锁定机制.用对偶计算系统减少冗余度。通过对偶计算可以仅检测计算的错误，并且通过知道操作的顺序可以提高系统效率。因此，我们已经实现了以下双重计算系统的并发控制机制。首先，在两个系统中的一个系统中执行操作，并且该系统将有效操作的序列发送到另一个系统。其次，另一个系统执行第一阶段发送的操作序列。3.从失败状态中恢复。通常情况下，我们使用的恢复机制中使用的最严重的故障的所有类型的故障，以简化恢复算法。我们通过根据故障类型选择恢复机制来提高系统效率。少

It is very important to implement highly available systems which make users unaware of temporal stoppages generated by software errors and hardware failures. Generally, hardware mechanisms are used to obtain highly availabilities. By using software techniques, following advantages are obtained.1. Temporal stoppages caused by software bugs can be handled.2. We can construct highly available systems economically because no special hardware are required.As there are some overheads which are required to synchronous redundant computation and to compare the result of each computation, the system efficiency is dropped. In this research, we have developed the concurrency control mechanism to improve the efficiency of the highly available system by using redundant computations. The following shows our results.1. Concurrency control for redundant computations. We recuce the overhead to synchronize redundant computations by executing redundant computations asynchronously as much as possible. We h … More ave shown that deadlocks and rollbacks are reduced by executing sequence of operations in the later system in the same order which is executed in the faster system, and we have developed Multi-Wait Two- Phase Locking Mechanism by extending this idea.2. Reduction of the degree of redundancy by dual computation systems. It is possible to detect only errors of computations by dual computation and the system efficiency can be improved by knowing the sequence of operations. Thus we have implemented the concurrency control mechanism for the following dual computation system. First, operations are executed in one of two systems and the system sends the sequence of valid operations to another system. Second, another system executes the sequence of operations sent in the first phase.3. Recovery from the failure state. Normally, wer use recovery mechanisms used in the most serious failure for all type of failures in order to simplify the recovery algorithms. We have improved the system efficiency by selecting recovery mechanisms depending on type of failures. Less