FEATURES OF DESIGNING SOFTWARE DISTRIBUTED SYSTEMS ARCHITECTURE

Abstract

A pressing problem in designing distributed software systems is that they must operate stably under high load conditions when thousands of users want to receive certain resources provided by system services. To ensure high availability and stability of highly loaded software services, they are deployed in managed multiprocessor distributed systems (clusters). These kinds of resources have a high value, so in practice, various cloud platforms (Google Cloud, Amazon, Azure, etc.) are most often used, which provide these resources, charging only for the time of direct use of them. Services must be able to fully utilize the provided resources during data processing, so they must be designed using special architectural solutions. Therefore, the purpose of this research is to theoretically investigate the architectural solution features when designing software services of distributed systems.

Materials and Methods. The paper presents a theoretical research design of features of distributed software systems architecture which is based on the analysis and comparison of facts obtained from scientific sources and the author's practical experience as well.

Results and Discussion. The article shows that Kubernetes is one of the main software solutions designed for deploying software applications in parallel distributed systems. It was established that microservice architecture is the optimal architectural solution for designing software services of distributed systems, given the specifics of deploying software systems under Kubernetes management.

Conclusion. A multithreaded design must be used for effective scaling distributed software system under high load. However, performance improvement occurs only if the parallel algorithm uses parallel rather than concurrent multithreading mode. The concurrent mode of thread operation is advisable to use only in the case of blocking operations when some threads are in a waiting mode. Synchronous architecture has a good performance, but there are limitations associated with blocking threads until the results of client requests are received. The asynchronous model allows for a larger number of requests than the synchronous one but requires a fully asynchronous API when working with external services, and is also more difficult to debug the service and fix errors.