随着Internet技术的广泛应用，大规模的信息发布系统逐渐得到使用。传统的同步通信模型由于其紧耦合性的特点，难以适用于这种大规模、异步和多点通信的需求，而发布订阅系统恰好能满足这些要求，因此应用越来越广泛。

本文在P2P网络模型基础上提出了自适应发布订阅系统架构，并针对该架构中负载均衡和延迟这两个关键问题开展深入研究。

首先，针对负载均衡问题，提出了通过使用订阅条件的覆盖、合并和绑定的订阅条件路由算法，该算法根据设定的参数合并概率和绑定次数将订阅条件分配到普通客户端节点上去，减轻代理节点的负载。还可以根据具体应用需求来调整参数来实现发布订阅系统的自适应性。针对延迟问题，本文提出了由Overlay层和网络层共同组成的事件双层邻接转发树算法。考虑到普通节点上保存的订阅条件数量较少，且订阅条件之间的相似程度高，使用订阅条件交集策略来减少匹配工作量以加快将事件路由到感兴趣的订阅者。并基于上述的方法实现了自适应发布订阅系统APSA（Adaptable Publish Subscribe Architecture）。

其次，通过分析发布订阅系统中使用纯P2P网络模型Pastry的不足，针对邻近路由导致延迟增大这一问题，设计了层次式P2P网络模型LDN（Low Delay Network），并基于LDN构建了新的发布订阅系统架构APSAX（Adaptable Publish Subscribe Architecture Extension）。

最后，本文分别对APSA和APSAX在仿真平台上进行了实验，从负载均衡、平均延迟和网络流量耗费三方面进行了对比分析。实验结果表明，APSA与目前的通用系统架构相比，能够实现可调的负载均衡，对于负载最大的代理节点，可以减轻近60%的负载，且延迟能够得到合理的控制。而基于LDN构建的APSAX，不管是延迟的最大值还是整体分布，相比于APSA都进一步降低，延迟最大值降低了近40%，充分证明了本文研究成果可以在保证负载均衡效果的前提下，实现降低延迟的目的，更能够适用于实际需要。

With the wide application of Internet technology, large-scale information releasing system gradually becomes popular. It is difficult for traditional synchronous communication model to apply to the needs of this large-scale, asynchronous and multi-point communications for its tight coupling characteristics. However, the publish-subscribe system is widely used for it meets these requirements.

    This paper proposes a publish-subscribe system architecture based on a P2P network and gives further study of load balancing and delay.

    Firstly, for load balancing problem, a routing method of  subscription covering, merging and binding is used in application layer. The method will assign a part of subscriptions to ordinary node based on the parameters and reduce the load on the proxy node. In addition, this method could make the system adaptable by using different values for parameters. For the delay problem, this paper presents a double-layer including overlay layer and network layer event forwarding adjacency tree algorithm and uses the subscription intersection strategy ,which is based on the characteristics of subscriptions, to accelerate the event routed to the interested subscribers. And then, the paper implements an adaptable publish-subscribe system based on the above-mentioned methods.

    Secondly, the paper analyzes the shortage of pure P2P network model Pastry in APSA and introduces a hierarchical P2P network model LDN(Low Delay Network). And then, it brings in a new publish-subscribe system APSAX(Adaptable Publish Subscribe Architecture Extension) which could greatly reduce the latency of event with good load balancing situation.

    Finally, this paper takes comparative analysis for APSA and APSAX on simulation experiment platform from load balancing, average delay and network traffic cost. Experimental results show that, APSA is adjustable in load balancing, which reduces almost 60% load from the view of maximum, and the delay can be controlled. What's more, APSAX reduces the latency for almost 40% from the view of maximum which fully proves the research of this paper could reduce delay on the promise of load balancing and meets actual requirements better.