Reliable multicast communication for parallel computation in distributed environments
Date
1990
Authors
Liu, Frances Fang
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Multicasting is a useful and efficient communication method for distributed and parallel applications. This thesis is concerned with the design and implementation of reliable multicast interprocess communication (IPC) mechanism for supporting distributed computations in an environment where certain types of failure could occur.
The need for reliable multicast communication mechanism arises in applications that are distributed to achieve parallel processing, resource sharing, data availability and reliability. Many of these applications require different levels of reliability when using multicast communication. Although most local area networks support broadcast communication, such broadcast facilities are rather low-level, unreliable and cannot be fully utilized by application programs.
This thesis presents a high level, decentralized protocol for achieving reliable multicast communication in distributed environments. The protocol guarantees that messages are received by all the operational receivers or by none of them(full delivery and atomicity). It also ensures that the messages sent from the senders will be delivered in the same order to all the receivers(global ordering).
The protocol introduces the simple but powerful concept of current status vectors which is attached to every outgoing message, together with a logical timestamp and a sequence number. These mechanisms enables each node to ensure atomicity and global ordering independently and in a distributed manner.
The proposed reliable multicast communication protocol has been implemented in REM(Remote Execution Manager). Performance data for benchmarks using reliable multicasting is compared with those obtained under unicast communication.