Parallel recovery method in shared-nothing spatial database cluster system

Shared-nothing spatial database cluster system provides high availability since a replicated node can continue service even if any node in cluster system was crashed.However if the failed node wouldn't be recovered quickly, whole system performance will decrease since the other nodes must process the queries which the failed node may be processed. Therefore the recovery of cluster system is very important to provide the stable service. In most previous proposed techniques, external logs should be recorded in all nodes even if the failed node does not exist. So update transactions are processed slowly.Also recovery time of the failed node increases since a single storage for all database is used to record external logs in each node. Therefore we propose a parallel recovery method for recovering the failed node quickly.     

Parallel recovery method in shared-nothing spatial database cluster system

Shared-notliing spatial database cluster system provides high availability since a replicated node can continue service even if any node in cluster system was crashed. However if the failed node wouldn’t be recovered quickly,whole system performance will decrease since the other nodes must process the queries which the failed node may be processed. Therefore the recovery of cluster system is very important to provide the stable service. In most previous proposed techniques,external logs should be recorded in all nodes even if the failed node does not exist. So update transactions are processed slowly. Also reccvery time of the failed node increases since a single storage for all database is used to record external logs in each node. Therefore we propose a parallel recovery method for recovering the failed node quickly.     

A recovery method using recently updated record information in shared-nothing spatial database cluster

A shared-nothing spatial database cluster is system that provides continuous service even if some system failure happens in any node. So,an efficient recovery of system failure is very important. Generally,the existing method recovers the failed node by using both cluster log and local log. This method,however,cause several problems that increase communication cost and size of cluster log. This paper proposes novel recovery method using recently updated record information in shared-nothing spatial database cluster. The proposed technique utilizes update information of records and pointers of actual data. This makes a reduction of log size and communication cost. Consequently,this reduces reccvery time of failed node due to less processing of update operations.     

A recovery method using recently updated record information in shared-nothing spatial database cluster

A shared-nothing spatial database cluster is system that provides continuous service even if some system failure happens in any node. So, an efficient recovery of system failure is very important. Generally, the existing method recovers the failed node by using both cluster log and local log. This method, however, cause several problems that increase communication cost and size of cluster log. This paper proposes novel recovery method using recently updated record information in shared-nothing spatial database cluster. The proposed technique utilizes update information of records and pointers of actual data. This makes a reduction of log size and communication cost. Consequently, this reduces recovery time of failed node due to less processing of update operations.     

A recovery method using recently updated record information in shared-nothing spatial database cluster

A shared-nothing spatial database cluster is system that provides continuous service even if some system failure happens in any node. So, an efficient recovery of system failure is very important. Generally, the existing method recovers the failed node by using both cluster log and local log. This method, however, cause several problems that increase communication cost and size of cluster log. This paper proposes novel recovery method using recently updated record information in shared-nothing spatial database cluster. The proposed technique utilizes update information of records and pointers of actual data. This makes a reduction of log size and communication cost.Consequently, this reduces recovery time of failed node due to less processing of update operations.     

机群系统中矩阵的并行QR分解算法

随着高速网络技术的快速发展，机群系统已经成为并行计算的主要平台，由于它的高通信延迟，某些在并行机上实现的细粒度并行算法己不适合在该环境下运行，为此有必要研究它们在机群系统中的并行实现。基于这一点，本文对矩阵的QR分解提出了一种新的任务划分策略，并由此得到了它的一种粗粒度并行算法。实验结果表明，设计的并行算法在机群系统中具有较高的加速比。     

混合动态数据库集群的并行空间连接优化算法

传统算法数据划分冗余度和倾斜度高,无用连接数据多,降低负载均衡性,对整体效率产生不好的影响,不适于实际应用。为此,面向混合动态数据库集群提出一种新的并行空间连接优化算法。采用网格划分法对数据进行划分,依据空间数据划分结果获取数据分布状态,计算节点按照数据分布状态得到候选集。通过平面扫描形成若干子空间连接的子任务,利用构建节点花费模型,依据花费模型对并行空间连接所需的平均节点访问个数进行评估,把候选任务集分配至不同计算节点,在不同节点执行并行空间连接操作。通过边界过滤策略,删除不可能有结果的元组,提高效率,增强算法的实用性。实验结果表明,所提算法适于实际应用,效率高。     

Oracle8i数据库系统的备份及恢复

介绍了0racle8i数据库中的3种备份方法：导入／导出备份、冷备份、热备份，结合3种方法应用提出了自启动备份策略，最后针对实际应用中出现的问题给出了相应的恢复策略。     

大规模分布并行计算系统容错与恢复技术

当前,拥有超级计算能力的计算机系统通常是大型商用系统形成计算机集群.与所有的分布式系统一样,这些系统通过独立的计算机硬件协同合作共同实现超级计算的能力.然而在拥有超级计算能力的同时,集群中的任何一个组件随时都可能失效,从而导致错的输出.为了提高集群在系统出现故障的情况下的鲁棒性,许多容错技术已经被设计和实现,用以处理各种类型的系统故障.本文对各种现有的容错技术进行了总结归纳,以便在此基础之上进行进一步的研究从而适应当前环境下的系统容错.     

An on-line scaling method for improving scalability of a database cluster

The explosive growth of the Internet and database applications has driven database to be more scalable and available, and able to support on-line scaling without interrupting service. To support more client's queries without downtime and degrading the response time, more nodes have to be scaled up while the database is running. This paper presents the overview of scalable and available database that satisfies the above characteristics. And we propose a novel on-line scaling method. Our method improves the existing on-line scaling method for fast response time and higher throughputs. Our proposed method reduces unnecessary network use, i.e. , we decrease the number of data copy by reusing the backup data. Also, our on-line scaling operation can be processed parallel by selecting adequate nodes as new node. Our performance study shows that our method results in significant reduction in data copy time.     

并 行 文 件 系 统 WPFS

介绍在Windows NT平台上研制的并行文件系统WPFS.     

在Linux下构造并行计算平台及开发并行程序

L inux是一个非常优秀的操作系统,其上的PVM也是目前最流行的并行编程语言之一.借助L inux构建并行计算系统,以及在L inux平台下开发PVM并行程序.     

机群环境下有限元的并行直接解法

在机群系统下进行有限元并行算法的研究是工程领域的前沿课题之一。本文详细分析了有限元的并行直接解法,解决了如何在机群环境下并行计算单元刚度矩阵和总刚度矩阵,并且按行或按列穿插存放在各个处理机局部存贮器中,并给出了相应算法。     

PC机群上FOX乘法的并行实现

分析了并行计算技术和机群系统的特点,提出了PC机群上FOX乘法并行实现的算法原理,并提出了改进算法.数值结果表明改进的算法收到较好的效果.     

基于集群系统的并行关联规则挖掘策略

文章在分析已有并行关联规则挖掘算法的基础上,讨论了多处理器系统中影响并行关联规则挖掘算法性能的主要问题。提出了多数据源在集群系统中的分布策略、在集群系统中进行并行关联规则挖掘的过程与策略、挖掘过程中并行计算的模型与方法以及规则的合并机制。     

求解三对角方程组的并行追赶算法

通过分析影响算法的并行效率的主要因素,根据分而治之策略中的分块思想提出了一种求解三对角方程组的并行追赶算法。然后在机群系统中,MPI环境下实现了该并行算法,对并行算法的加速比和效率与原串行算法进行了比较,结果表明此算法有较高的计算效率。     

An on-line scaling method for improving scalability of a database cluster

The explosive growth of the Internet and database applications has driven database to be more scalable and available, and able to support on-line scaling without interrupting service. To support more client‘s queries without downtime and degrading the response time, more nodes have to be scaled up while the database is running. This paper presents the overview of scalable and available database that satisfies the above characteristics. And we propose a novel on-line scaling method. Our method improves the existing on-line scaling method for fast response time and higher throughputs. Our proposed method reduces unnecessary network use, i.e. , we decrease the number of data copy by reusing the backup data. Also, our on-line scaling operation can be processed parallel by selecting adequate nodes as new node. Our performance study shows that our method results in significant reduction in data copy time.     

一种集群系统的实现

计算机网络技术的飞速发展使得对网络服务的要求越来越高.如何针对大量客户群体提供可靠而有效的服务,既保证服务质量又减少成本.集群系统使得多台普通服务器可以联合工作,从而达到较高的运算能力和可扩展性.作者在参考了上述技术之后,针对网络服务这一专用领域的集群系统进行了研究,提出一种优化的方案,并详细介绍了它的实现细节.     

阶段并行模型上的并行算法设计

阶段并行模型是BSP（Bulk Synchronous Parallel）模型的改进,它更接近于表述实际的机器行为,同时具有编程简单、独立于体系结构和执行性能可预测等特点。NOWs正成为并行计算领域的的一个新的发展热点,以太网构成的微机集群系统是NOWs的一种重要实现形式,研究了群集系统中阶段并行模型上的并行算法设计,以FFT算法为例,进行了设计和分析,并给出了测试结果。     

一类延迟微分方程的并行Rosenbrock方法

针对刚性大系统,根据实际数值仿真和科学计算的需要,提出了一类并行Rosenbrock方法.该方法将不同级分配到不同的处理器上同时计算,以提高计算效率.将其用到一类延迟微分方程上,并对其稳定性及收敛性进行讨论.该方法不需要迭代,具有良好的稳定性.