【作者】 徐浩；

【导师】 肖卫东；

【作者基本信息】 国防科学技术大学 ， 管理科学与工程， 2009， 硕士

【摘要】 在分布式网络化作战环境中引入对等(Peer-to-Peer,P2P)网络,并对战场资源进行冗余存储,可以避免将这些战场资源集中存储在少数几个存储量大、处理能力很强的服务器中,以至于使其成为敌方的重点打击目标。因此,研究面向数据抗毁性的对等网络数据冗余存储策略,有着极其重要的意义。当节点在不可预测的情况下失效时,现有的评价指标不能准确评价对等网络中数据被访问的概率,因而需要引入一种与节点有效性无关的评价指标来进行描述,由此本文提出了数据抗毁性这一概念。随后,本文从全复制、分块复制和有固定中心节点的分块复制三个角度对数据抗毁性进行建模分析,并进一步讨论了数据恢复机制对数据抗毁性的影响。基于本文所进行的数据抗毁性的描述,在相同的存储开销下,分块复制策略下的数据抗毁性在失效节点数量较少时要高于全复制,而当失效节点数量较多时,全复制策略下的数据抗毁性要更高。因此,本文设计了一个分块复制与全复制相结合的冗余存储策略,对数据进行分块存储的同时在网络中保留一个完整的数据副本,将分块复制与全复制的优势相结合。然而现有的数据存储策略在进行冗余存储时,网络中通常会有固定的中心节点进行这些冗余数据的管理,由中心节点主动发出探询消息,以检测网络中的副本数量、位置等信息。这些节点的存在方面简化了算法,但同时也造成了中心节点的瓶颈效应。而且通过本文描述,采用有固定中心节点的分块复制算法所获得的数据抗毁性要低于全复制和分块复制算法。因此,本文充分考虑到中心节点的瓶颈效应,提出了基于动态中心确认机制的数据存储策略DCDS(Dynamic Center-based Data Storage)及其改进策略IDCDS(Improved DCDS),将数据副本的确认模式改为数据块保存节点主动确认。利用基于DHT的结构化P2P网络的特性,由数据块保存节点将数据D的确认信息发往该数据的key值管理节点,并由该数据的key值管理节点来负责检测网络中数据块的数量。在这种模式下,即使节点失效,根据基于DHT的P2P网络的特点,会有新的节点来管理该key值,那么数据块的确认信息就会发至这个新的节点上,从而确保数据不会随着确认节点的失效而丢失。仿真实验表明本文提出的基于动态中心确认机制的数据存储策略在网络节点不断失效情况下,对提高网络中可用数据的数量是有效的。更多还原

【Abstract】 A peer-to-peer(P2P) networks and redundant storage strategy in a distributed networked operations environment can avoid the centralized storage in the few and high capacity of the host which can easily become the focus of the enemy’s attack. Therefore, data invulnerability oriented redundant storage in peer-to-peer networks is of great significance.When node failure appears in an unpredictable way, the existing evaluation is not accurate to the probability of data being accessed. Hence, a new evaluation, which is irrelative about the node availability, is introduced. After that, this dissertation models the data invulnerability through three perspectives:replication, blocking replication, blocking replication based on a fixed central point. Besides, data reconstruction strategy is discussed.Based on the data invulnerability proposed in this dissertation, blocking replication strategy gains higher data invulnerability than replication strategy when the data storage costs are the same and there are few node failures in the networks. But the replication strategy gains higher data invulnerability than blocking replication strategy when the failure increased. So, the dissertation presents a mixed redundant storage strategy, which stores a full replica in the networks when useing the blocking strategy. However, there is always a central point in current storage strategy which managing the redundant data, such as sending probe message actively and detecting the number of replica. On the one hand, these central point can reduce the complexity of the storage strategy. On the other hand, they become the bottleneck of the efficient storage, and the blocking replication based on a fixed central point gains less than the replication and blocking replication. Thus, a dynamic center-based data storage(DCDS) strategy and a improving-DCDS(IDCDS) strategy are proposed. It changes the passive confirmation of blocking data’s storage node into the active confirmation. The blocking data’s storage node send confirmation message using distributed hash table(DHT) to the node which manage the blocking’s key value. And the latter one will detect the status of the data. When it fails, the confirmation message will be sent to new comer under the DHT-based P2P networks. Therefore the data won’t be lost with the node failure.Simulation results show that the DCDS strategy and the IDCDS strategy can incerase the number of data in networks efficiently on the premise of node failure continuously.更多还原