Research On Key Technologies For Supporting Content-based Publish/Subscribe

Distributed object-oriented communication middleware aims at solving the problem of interoperability among distributed software in heterogeneous network environments, which can hide the detail of network communication and improve the portability. So it has been applied in many fileds and has become the major technology for implementing distributed computation. With the rapid development of Internet, the pattern of software has shifted from closed, the local familiar user groups and relative static tight-coupled form to open, public accessing and dynamic coordination loosely-coupled form. The single synchronous communication mechanism which is provided by traditional object-oriented middleware based on RPC mechanism can't satisfy the dynamic characteristic of Internet. To coupe with this challenge, it is necessary to provide an asynchronous and loosely coupled communication paradigm for building distributed application at top layer.Publish/Subscribe is a communication paradigm based on event. Compared with asynchronous RPC, distributed shared memory and message queuing, it provides complete decoupling in space, time and control flow for participants. So it can satisfy the requirement of loosely coupled communication. It has gotten broad attention and deep research.Publish/Subscribe can be divided into channel-based, topic-based and content-based system according to the notification selection mechanism. Channel-based and topic-based publish/subscribe is simple, but content-based publish/subscribe has more powerful, more flexible and broad abroad adaptive character. So content-based publish/subscribe is one of the basic infrastructure to construct new pattern distributed applications in the future. But compared with topic-based publish/subscribe, content-based publish/subscribe is more complex. In order to grow into a practical infrastructure to support large scale distributed applications in deed, there are many key technologies to be solved yet.We prepose some new ideas and methods in some key technologies on content-based publish/subscribe.Firstly, research on routing algorithm. Subscription covering has been proposed to reduce the size of routing table in intermediate brokers in the publish/subscribe networkwe and to decrease the time of event matching. We distinguish routing based on subscription covering into optimal and suboptimal policy. We propose strict subscription-cover based routing (SSCBR) and relaxed subscription-cover based routing (RSCBR). The former uses optimal subscription covering policy, whose goal is to maintain the least covered subscriptions. The latter uses suboptimal subscription covering policy, which maintains more covered subscription to balance the cost in memory, time and network traffic. On the other hand, we propose a necessary and sufficient condition of correct routing table configuration when using SSCBR idea. We prove the correctness of SSCBR and RSCBR routing algorithm based on the necessary and sufficient condition.Secondly, research on reliable event message transfer. We suppose a two layers reliability model to assure reliable event transfer, which are reliable P/S protocol layer (RPSL) and reliable stream transfer layer (RSTL). The RPSL is aware of the format of P/S message. The RSTL doesn't concern with the format of RPSL message, in which RPSL message is regarded as data stream. The two layers can provide support for reliable event transfer individually. User can use any one of two layers or use them both. In RPSL layer, based on combined push and pull method using gossip algorithm, we present a event-loss detecting and recovery algorithm which supports subscription covering routing. The algorithm benefits the advantage of less overhead from pull method, and overcomes the disadvantage of pull method which can't detect event loss in some scenarios. The RSTL supports reliable message stream transfer in point to point through adaptive to link rate and break-and-renewal mechanism.Thirdly, research on supporting mobile computing. For the temporary disconnection of subscriber, we present a fast reconnection algorithm. This algorithm can assure at-most-once semantics. For event migration when durable subscribers hand off, we analaye the validity of existing algorithms and their dependence on routing algorithm. We present two enhancements, named pre-divert and pre-recover. Pre-divert method forwards events to new borker in advance. Pre-recover method recover events from new broker and its neighbor when event migration algorithm is synchronizing. These two methods decrease the number of event that needs to be migrated, so they reduce the time of event migration procudure.Fourthly, we implement a notification service protype named Densa. Densa has more flexible ability of configuration and scalability. Applications can configure the policy of routing, the QoS of reliable event transfer, so as to satisfy their requirement. Densa can support many communication protocols through multiple protocols management framework, which can be used under complex appication enviroment.To sum up, we present elementary solutions for some key technologies in content-based publish/subscribe. We expect that our investigation can bring edification in theory research and reference in pratical application.