探讨查询分布式数据分片存放策略.docVIP

探讨查询分布式数据的分片存放策略 1．1数据流查询处理相关的概念定义以及假设说明 1.1 data stream query processing related concept definition and assumption 　　集中式数据流查询处理的体系结构由两部分构成，即查询计划生成子系统（FRONT－end）以及查询执行子系统(BACK)。其中两部分与关系数据库系统相比均有较大的区别。查询执行子系统如图1所示。 Centralized data stream query processing architecture consists of two parts, namely the query plan generated subsystem (FRONT end) and query execution subsystem (BACK). Part two have significant difference compared with relational database system. Query execution subsystem is shown in figure 1. 　　 　　通过这种散列，将系统当前的所有查询映射到节点空间，然后由该节点上的查询处理器完成到达的查询。 Through this hash, to all queries the current system is mapped to a node space, then finished by the query processor on the nodes to query. 　　b)查询内并行处理方式。在系统的范围内，由操作符、输入均输出记录队列、维持操作符状态的大纲信息构成网状结构。 B) the query in parallel processing mode. Within the scope of the system by the operator, the input output record queue, maintain the state of the operator outline information constitutes a mesh structure. 　　c)命名发现机制。参与查询处理的节点有全局惟一命名participant（如IP地址等）。当在一个节点上面定义一个新的流模式、数据流、操作符，这些实体均隶属于其命名空间。该实体可以采用下面的命名方式：（participant，entity－name） 。为了了解系统中数据流模式的定义、系统中的数据流、数据流的到达（存放）位置、系统中哪一部分查询执行，就要考虑在catalog中存放必要的数据。其中catalog信息是通过在DHT下分布式存储的，前面已经分析了catalog信息的存储问题。 C) named discovery mechanism. Involved in query processing node has a globally unique name of the participant (e.g., IP address, etc.). When a node defined above a new flow model, data flow, operators, these entities are belongs to the namespace. The entity can be used in the following way: (the participant, the entity - the name). System data flow pattern in order to understand the definition, system data flow, data flow in the reach (store) position, which part of the query execution system, consider the  as necessary in the catalog data. The catalog information is through distributed storage under the DHT, as already analyzed the catalog information storage problem. 　　系统中对每一个数据流、每一个查询、查询中的算子、算子大纲、节点间输出队列均有惟一的命名。查询处理器位于DHT之上。同查询相关的数据粒度限定为数据流、输入数据源（记录集）、节点间传输数据队列、算