System and method for monitoring system availability based on statistical method

Abstract

The invention proposes a system and method for monitoring the system availability based on a statistical method, and the method comprises a system-to-system service call log module, an alarm thresholdanalysis module, a warning alarm module, and a monitoring alarm display module. The system regularly performs the statistical learning of historical data through the collection of call logs between systems, and obtains the general performances of each system. The system performs the analysis of the data in a last time unit t, and recognizes whether the number of current errors of each system is abnormal or not, whether the error rate between system-to-system call is abnormal or not, and whether the usability of all systems and examples of the system are abnormal or not. The system marks the abnormal systems and the call relation between the abnormal systems on a system topological graph in an alarm mode. During the displaying of the alarm information, the system states, the call states between systems and the states of the system services and examples are displayed on the topological graph, so as to achieve the quick locating of systems with problems when systems in a large area haveproblems.

Description

Technical field [0001] The invention belongs to the technical field of software system monitoring, and relates to a system and method for monitoring system availability based on a statistical method. Background technique [0002] Internet companies generally contain a large number of application systems. In addition to open websites and APPs, there are also many internal application systems to support the operation and management of the enterprise. In general, there are more complicated calling relationships among internal application systems, and the functions that one system provides to another system are called services. The application system availability monitoring industry generally adopts the following methods: [0003] Method 1: Use tools such as zabbix to monitor certain indicators of the system server, such as the number of web system processes / threads, CPU load, available memory, the number of http abnormal status codes, and request response time. When the indicator ex...

AI Technical Summary

Problems solved by technology

[0006] 1. The thresholds in method 1 and method 2 need to be set manually. The thresholds of different systems vary greatly, and the thresholds of the same system in different periods are also completely different. The setting and maintenance of thresholds have a lot of workload

In actual operation, the trial and error method is generally adopted, that is, the threshold is relaxed after false positives, and the threshold is tightened after false negatives, so that the false positive rate and false negative rate are very high

[0007] 2. The monitoring of method 1 can only partially reflect the availability, but cannot be used as an actual availability indicator. The detected abnormalities do not mean that the system availability is reduced, and when the system is unavailable, they are not all reflected in these monitoring indicators.

[0008] 3. The monitoring of method 2 directly reflects the usability, but as a sampling method, the number of samples is small and the coverage is narrow. It can only monitor read operations and is less used for write operations.

[0009] 4. When the system is large and complex, the above two monitoring methods have too many indicators, a large number of alarms, and loud alarm noise, which will affect the judgment and location of the problem

[0010] 5. When a new system is launched, a new service is launched, or system and service deployment changes, the above two monitoring methods require manual maintenance of monitoring items, and are not suitable for systems with automatic fault switching and dynamic expansion of service capabilities

[0011] 6. When performing error rate monitoring and alarming, the threshold method often causes false alarms. For example, when the error rate requirement does not exceed 1%, if only one operation occurs and fails (error rate 100%), it will alarm, but in most cases no warning

[0012] 7. When multiple systems in a complex system cluster fail at the same time, it is difficult to quickly locate the system that actually failed, and you can only grab the beard and eyebrows, wasting precious time

Images