Abstract
This work is focused on a number of standard communication patterns of distributed system nodes via messages. Certain characteristics of modern practically applied communication systems are considered. The conclusions are based on the practical development of collective communication strategy processing services and the theoretical basis drawn in the course of testing a number of distributed system prototypes. Development trends of service oriented architecture in the field of interservice communications are considered, including the development tendencies of AMQP and ZMTP protocols.
Problems arising during the design and development of such systems from the horizontal scaling standpoint are specified. The problem of long term control is highlighted in the course of considering issues of data consistency between nodes, availability and partition tolerance. The process of changing workload distribution in a horizontally scaled system is described and issues of fault tolerance of the system in general and its nodes in particular are raised. A way of workload scaling by means of defining an evaluation criterion of node load determined by the system’s business logic and not by the characteristics of the communications level is offered. The efficiency of this approach is shown, with long term control systems used as an example.