Dynamic Thresholding Detects “Abnormal” Resource Utilisation and Patterns
Dynamic thresholding makes use of self-learning analytics to understand the “normal” range of VM resource usage in a virtual environment. Because every environment can be vastly different, these analytics observe consumption of resources over a period of time to understand usage. For example, if a VM displays high CPU utilisation on the same day each week, these analytics “learn” that this is a usual occurrence and will consider this to be the baseline utilisation, dynamically setting warning thresholds differently for this day. As a result, this VM would be considered to have a high CPU utilisation performance issue only if the CPU utilisation is vastly higher than usual for this specific day of the week. Through this method, “abnormal” behaviour for resource usage is dynamically determined and false positives can be removed for behaviour that is shown to be typical.
While dynamic thresholding based on self-learning analytics is valuable for analysing virtual environments, multiple analytic types are required to detect all sorts of virtualisation issues. Because dynamically set thresholds are specific to each VM’s observed resource usage, issues that exist while a baseline is being established will not be considered problematic. Additionally, many issues cannot be detected with dynamic thresholds, such as memory swapping, accelerated storage utilisation and high disk latency as they require metric-specific static threshold alarms. Virtualisation management systems which rely solely on dynamic thresholding will be unable to detect these and many other kinds of issues. VKernel’s approach is to build and deploy the right types of algorithms to maximise accurate analysis of virtual environments.
With vOPS Server Enterprise 6.6.2’s new feature set, dynamic thresholding adds precision in determining which resource usage patterns are normal or abnormal, in VM CPU, memory, storage and disk I/O utilisation. This is in addition to other analytic types existing within the vOPS product to detect VM performance issues.
“Dynamic thresholding enabled through self-learning analytics are unique in its ability to detect abnormalities in resource utilisation specific to the virtual environment,” says Bernd Harzog, The Virtualisation Practice. “The alarms can only help system administrators be more proactive in avoiding performance issues. They are now alerted when an environment is beginning to show abnormalities in resource usage that could potentially turn into performance issues and result in environment downtime.”
Dynamic Thresholding Bolsters Multi-Analytic Approach in vOPS Server Product Line
vOPS Server Enterprise 6.6.2 features dynamic thresholding for VM resource utilisation by enabling the IntelliProfile self-learning analytics engine. IntelliProfile is a mature technology featured in other Quest products to detect abnormalities in usage in applications such as Microsoft SQL Server. Dynamic thresholding will complement existing analytic types within vOPS Server Enterprise such as threshold-based alarms and accelerated growth alarms to expand the total number of issue types that can be detected by the vOPS Server product line.
“We are pleased to integrate dynamic thresholding in vOPS Server Enterprise based on the same IntelliProfile engine that has been used and improved over many years in other Quest monitoring solutions,” says Mattias Sundling, Product Evangelist and vExpert, VKernel. “Adding this capability to the multiple issue detection analytic methods that currently exist within the vOPS Server product line will allow our customers to get the most sensitive VM performance issue detection system available on the market.”
Tags: Server Virtualization, VMware