As systems become more important in business, expectations for stable system operation are increasing. Predictive detection, which detects signs of system failure, can be said to be a method that contributes to stable system operation.
In this article, we will introduce the overview, technical background, and main methods of “predictive detection,” which can be said to be useful in system operation management.
table of contents
- “Predictive detection” attracting attention in system operation management
- The importance of dealing with failures “before they occur”
- What is predictive detection?
- Technical background of predictive sign detection
- What does predictive detection achieve?
- Prevention of failures in advance
- Predictive failure detection and alert reduction
- Main sign detection methods
- Dynamic threshold setting
- Capacity prediction
- Analysis of periodicity
- Achieve early detection by introducing monitoring tools
- What is “LogicMonitor”, an integrated operation monitoring tool that invests in advanced predictive detection functions?
- Dynamic threshold setting
- Forecasting future capacity
- Reduce noise alerts
- summary
“Predictive detection” attracting attention in system operation management
In recent years, the keyword “predictive detection” has gained attention in system operation management. What kind of concept is predictive sign detection? I will explain it below.
The importance of dealing with failures “before they occur”
Against the backdrop of trends such as DX, the scope of system use in business continues to expand. In business, the existence of systems has become a part of competitiveness, and system outages are directly linked to business shutdowns and a decline in corporate profitability.
Under these circumstances, the standards required for system operation management are also rising. In addition to the speed of recovery from failures in the event of a system outage, emphasis is now placed on efforts to prevent the system from stopping in the first place.
What is predictive detection?
Against this background, efforts are being made to ensure stable system operation. One example of this is predictive detection, which allows us to detect small changes and take proactive measures before they have a major impact, such as a system outage.
Predictive detection in IT system operation management is a mechanism for detecting signs of failure in advance by referring to past data such as load status, operating status, and performance status collected during system operation.
Normally, when monitoring system operations, we understand the occurrence of failures by checking the operating status and monitoring the thresholds of metrics information. On the other hand, predictive detection uses past data such as system performance to analyze patterns related to operating status.
Based on the analysis, we aim to understand the tendency for failures to occur and to detect signs of failures before they occur. As a result of the analysis, if a sign of a failure is detected, an alert will be issued. This allows proactive measures to be taken before an actual failure occurs.
Technical background of predictive sign detection
There are various methods for detecting signs, but one method is to statistically analyze past data such as system operating status.
For example, consider the case of realizing predictive detection based on CPU usage rate. When we statistically processed past CPU usage data and calculated the average value and standard deviation, we found that the average CPU usage rate was around 30%, and in most cases it was within the range of 0% to 70%. Masu. In this case, for example, if the CPU usage rate exceeds 70%, it can be determined that an exceptional event is occurring. By setting an alert to be issued at this time, you can recognize at an early stage that there may be a problem with the system.
There are also cases where AI-related technologies such as machine learning are used as a more advanced predictive detection method. For example, machine learning can be used to monitor whether events similar to abnormalities that have occurred in the past are occurring.
Furthermore, the use of AI to streamline and automate operational management tasks is called “AIOps.” AIOps is a term coined by Gartner, a major IT research company, in 2016, and is an abbreviation for “Artificial Intelligence for IT Operations.” AIOps can be said to be a way of thinking that leads to stable system operation while reducing the burden of system operation management.
What does predictive detection achieve?
What can be achieved by introducing predictive detection? Below, we introduce the main effects of implementation.
Prevention of failures in advance
Predictive detection allows you to catch the signs of failure before it occurs, allowing you to take action before the system stops.
For example, suppose you have detected a trend in the amount of resources used, such as CPU, increasing on a particular day of the week or during a specific time period. In this case, it can be assumed that some kind of problem is occurring in the processing being performed on this day of the week and at this time. When we investigated the processes being performed on the same day of the week and at the same time, it became clear that there was a problem with the behavior of a particular batch process, which was consuming a large amount of resources.
In this way, by understanding situations where abnormal behavior is occurring before a failure occurs and taking appropriate measures, you can avoid the worst-case scenario of a system outage.
Predictive failure detection and alert reduction
Furthermore, with advanced predictive detection based on the AIOps concept, it is also possible to detect signs of failure predictively.
Generally, system anomaly detection involves checking whether metrics collected through monitoring exceed a certain threshold, but it is difficult to set strict conditions for all situations that indicate a failure. . Failures may also occur due to conditions that have not been set up in advance.
By utilizing machine learning and AI functions to perform advanced analysis and learning of previously collected metrics data such as system operating status, we are also working to identify signs that failures are likely to occur, in addition to pre-set conditions. It will be possible.
AIOps can also be used to reduce unnecessary alerts.
Excessive alerting places a heavy burden on system operations administrators. If you end your work just by responding to unimportant alerts, you won’t be able to do other tasks. Additionally, when processing a large number of alerts, there is a risk of overlooking the cause of a failure.
Therefore, by using advanced functions based on the AIOps concept, we will perform the necessary and minimal alerting only for the fundamental problems that need to be addressed. This reduces the burden on system operations administrators, allowing them to focus on the necessary responses, leading to stable system operation and early recovery in the event of a failure.
Main sign detection methods
Below, we will introduce the main predictive detection methods.
Dynamic threshold setting
The optimal threshold value varies depending on the system environment and operating conditions. Generally, threshold settings for monitoring are often based on values during normal operation, but depending on the situation, a large number of alerts may be generated or signs of failure may not be detected.
Therefore, based on data on the past operating status of the system and performance data, we statistically analyze whether the system is operating stably or unstable, and dynamically set thresholds. This will ensure that you receive sufficient alerting.
Capacity prediction
For resources such as storage, which will become depleted as usage increases over a long period of time, it is necessary to anticipate when the capacity will run out. Therefore, technology is used to predict future capacity.
In capacity prediction, future resource usage is predicted based on past resource usage. For example, if your storage usage increases by an average of 5% every month, and your current usage is 60%, you can easily calculate that you will run out of storage space in 8 months.
Of course, in reality, the amount of storage used fluctuates due to seasonal factors and irregularities, so it is necessary to estimate the degree of safety, but this kind of regressive prediction makes it possible to make some assumptions.
Capacity forecasting allows you to predict in advance when additional IT investment will be required, which has the advantage of making it possible to implement planned initiatives, including securing a budget and making internal adjustments.
Analysis of periodicity
The operating status of the system may be cyclical depending on the season or time of day.
For example, in your company’s accounting system, payroll processing is performed on a specific day of the month, and at the end of the month, invoice issuance processing and acceptance inspection processing are scheduled.
In such cases, it may not be possible to accurately detect signs by simply analyzing past performance data.
In the example of the accounting system mentioned above, there will be a high load on the payroll calculation day and at the end and beginning of the month. Additionally, in a system where online processing by users is the main activity, the CPU usage rate during the day may average 40%, but at night it may average 20%.
By analyzing periodicity in this way, more accurate sign detection becomes possible. For example, by distinguishing between the expected CPU usage rate during the day and the expected CPU usage rate at night, it will be possible to detect irregularities in the system while avoiding excessive alerting.
Achieve early detection by introducing monitoring tools
Predictive detection has various advantages, but how can it be incorporated into system operations?
In recent years, system operation monitoring tools that incorporate predictive detection technology have appeared. By incorporating such tools, it is possible to detect early warning signs.
Functions related to early warning detection, such as the dynamic threshold setting and capacity prediction mentioned above, vary depending on the product. Therefore, the key is to select products that suit your company’s needs.
What is “LogicMonitor”, an integrated operation monitoring tool that invests in advanced predictive detection functions?
LogicMonitor is a SaaS-type IT integrated operation monitoring service that is equipped with functions based on the AIOps concept introduced in this article. By introducing LogicMonitor, you can realize the following advanced symptom detection.
Dynamic threshold setting
Accumulates and learns normal patterns regarding operating status based on past data. We understand trends that are likely to cause failures, such as when things deviate from normal patterns, and detect signs in advance in real time.
Forecasting future capacity
Predictive calculation of future usage status using AIOps function. You can predict the amount and timing of additional investment for the optimal IT system, improve investment efficiency and optimize the entire system without wasting anything.
In addition, it is possible to optimize IT resources such as servers and storage while maintaining a balance by continuing to maintain appropriate resources and performance based on past operational data, including sizing IT resources.
Reduce noise alerts
A mechanism is in place to notify only root cause alerts to the appropriate operations staff at the optimal time. Significantly reduce noise alerts related to the root cause of the bottleneck, resulting in faster MTTR.
With these functions, LogicMonitor can also reduce the burden on system operations administrators. It can be said that it is a product that contributes to streamlining and automating IT operations.
summary
In this article, we introduced the main methods and their effects regarding predictive sign detection. As systems become more important in business, it is becoming increasingly important to avoid system outages that lead to lost opportunities. By making good use of tools that enable predictive warning detection, it will be possible to contribute to the stable operation of the system.
