Метод визначення цільових показників продуктивності програмного забезпечення

Abstract

Development of the methods for the determination of the target indices for measuring the quality level , performance reliability is one of the important directions in the sphere of information technologies and, in particular, software engineering. Among the known methods we can distinguish the method of balanced management indicators, SMART-techniques, etc. GIST (Goals, Ideas, Step-Projects, Tasks) methodology it is the method of production planning, elaborated by I. Gilad former Google production manager. This methodology is aimed at reduction of the overhead cost for management, enhancement of the speed of development and creation of the products, which meet the requirements of the market and their subject area. Defining target performance indicators remains a pressing challenge for modern DevOps teams. They are frequently caught between two extremes: collecting thousands of time-series metrics, which inflates costs and alert noise, or, conversely, collecting too little data and missing performance regressions. Lightweight checklists such as the Four Golden Signals or RED/USE lower the entry barrier but remain static; they ignore business goals, service criticality, and lifecycle stage. The paper proposes to improve GIST (Goal Impact Stage Template) method for measuring software performance (GISTSP). Objective c of the study is to create and evaluate the method of target indices formation on the base of the methodology of metrics collection GIST. The suggested GIST method (Goal – Impact – Stage – Template) relies on a four-line “service passport” and a Coreplus-Plus metric library. By following a “collect only what you need” rule, GIST automatically generates Prometheus/Grafana configurations and alert rules. The approach was experimentally compared with Four GS, RED/USE, and AWS W-A on four testbeds (Web-API, Queue, Stream, Batch). Results show that required SLI coverage increased from 78 % to 92 %; daily metric volume decreased by 60 %; setup time was reduced by a factor of four, while the F₁-score for regression detection did not degrade. These findings indicate that GIST achieves a practic.