Evaluation of Scalability of Distributed Systems based on Algorithms for Iterative Task Execution with Coordination via Key-Value Storage

This paper presents the results of research aimed at evaluating the scalability of distributed fault-tolerant systems for iterative task execution. The study focuses on peer-to-peer distributed systems capable of executing tasks iteratively according to a schedule, where the nodes are coordinated through a key-value store. The operation of such systems is based on Recurrent Worker Service (RWS) algorithms. The primary focus is on studying the dependency of metrics such as iteration accuracy and resource utilization on the number of system nodes, including both application and storage nodes. The experimental environment is deployed in Yandex Cloud infrastructure and consists of four servers, one of which is low-performance and three are high-performance. Tests were conducted under three scenarios: application node scalability, storage node scalability, and simultaneous scalability of both node categories. During the tests, performance differences between the two modes of RWS algorithms were studied. The results obtained in this work can be used in the design and optimization of distributed systems that require high performance and fault tolerance.