Transis: Efficient Multicasting in a Partitionable Setting

资源类型: 资源大小: 文档分类: 上传者:文彦君


【标题】Transis: Efficient Multicasting in a Partitionable Setting

【作者】 James Atkinson 

【摘要】Introduction Transis is a multicast communication layer with high fault tolerance. It is a package that developers can use to simplify the task of communication between processes. Transis is ideal on networks with a low message loss rate; but computers and connections may go down for some period of time. The Transis project began in 1991 at the Hebrew University of Jerusalem. It combined principles from two prior distributed systems: Trans; developed at Santa Barbara University and ISIS; developed at Cornell [4]. Transis's highly efficient multicast protocol comes from the Trans project; while some its group membership principles are derived from ISIS. Process Group Communication is supported. This means that Transis allows the user to name groups of processes; and send messages to every process in that group by specifying the group name. This way the user does not have to identify the targets of messages explicitly or find the network routes to them [5]. Transis guarantees all-or-none semantics. This means that all of the active processes in a group will receive a message sent to that group [5]. Network Partitioning What truly distinguishes Transis from its predecessors is that it allows network partitions and merges without losing consistency. A network partition is when two parts of a network find themselves temporarily disconnected. The result is two separate networks unable to communicate. A programmer who takes advantage of the enhanced facilities of Transis can write applications that will not fail in the case of a network partition. Instead; each component will continue to operate independently until the connection is restored. Then; they can re-merge while maintaining consistency [2]. Partitioning is an unavoidable consequence of the wide area network; and being able to handle it is important to many applications.