Method and algorithm for automatic recovery of information services based on objective predictive monitoring data

Schemelinin Dmitry A.

This paper presents the results of analyzing the performance and evaluating the use of the RAM of virtual machines used for Java-based applications and the developed method and algorithm for automatic recovery of services of globally distributed computing systems (GDVK) based on objective monitoring data, as well as predicting anomalies for a given period of time. in the future using new forecasting methods. To analyze data on the performance of computing resources, production statistics were collected from the studied GDVK using the continuous monitoring system Zabbix, which is used to regularly receive counters from several VMs (Virtual Machine) and store performance data in database management systems (DBMS) SQL (Structured Query Language) for analysis and troubleshooting. The production servers under study are organized into pool groups, each of which performs specific business services. The purpose of the analysis was to develop control methods, performance assessments and automatic elimination of anomalies in the software (software) of software components built using the JEDI technology (Java Environment for Distributed Invocation). The new monitoring method takes into account the specifics of Java memory leaks in Internet applications (Memory Leak), is based on predictive models of anomaly detection and provides secure automatic recovery of cloud services. Unlike existing solutions, the developed method and algorithm for automatic restoration of GDVK services is based on objective monitoring data, as well as predicting an anomaly for a given period of time in the future using new forecasting methods and is universal and can be generalized to a wider class of studied metrics and monitoring systems. The method was implemented programmatically at RingCentral on the example of detecting Java memory leaks in Internet applications and automatically restoring services in order to achieve a world level of 99.999% availability in 24/7 mode.

Keywords:monitoring, big data, modeling, forecasting function, monitoring metrics, Java, cloud technologies.