Alert communication primitives above TCP

摘要

We consider communication primitives that can be executed by an application process to exchange messages with another application process over a TCP/IP network. A communication primitive is called alert iff it satisfies two conditions. First, if during any execution of the primitive no failure occurs, then the execution completes successfully. Second, if during any execution of the primitive some failure occurs, then the execution is aborted and the process that initiated the execution is informed of the failure. Clearly alert communication primitives are useful in designing reliable distributed applications. We argue that the send primitive over TCP is alert, but the receive primitive over TCP is not. Then, we propose three new receive primitives over TCP and show that each of them is alert. We also discuss how to implement these three primitives and compare their performance.