VMware performance is a top concern for many application administrators when a virtualization initiative begins and can become a quagmire for a virtualization administrator if an issue strikes. While some VM performance issues may occur unexpectedly, such as an unforeseen spike in web traffic that strains disk I/O for a web server, many conditions that can lead to VMware performance issues like upward trending VMware memory utilization can be detected in early stages. Nipping such problems in the bud can be accomplished by implementing a regular VM monitoring process.
Implementing a VM Monitoring Process
For starters, a data center must resolve to take the time to execute a VM monitoring process with regular environment "check-ups". Ideally, a VMware administrator will examine system performance on a daily, weekly, and monthly basis to evaluate trends in capacity availability and usage growth at different intervals:
- Daily Monitoring - Virtualization administrators can evaluate the last 24 hours of system operations to see if there have been usage spikes or other unusual conditions that could lead to immediate VM performance problems. In particular, peak resource utilization times such as intensive batch processes that are run at night, or the morning hours when an organization’s employees begin logging in to applications should be examined. Seeing how a virtualized environment reacts at high utilization periods can yield insights that identify virtualized infrastructure areas that are coming under stress. Such areas will be the first to succumb to capacity bottlenecks that can cause VM performance problems if VM resource utilization increases from regular usage growth.
- Weekly Monitoring – A more in-depth examination of the virtual environment should occur on a weekly basis to see if there has been any continued change or growth in utilization throughout the week. Trending changes over several days may be symptoms of underlying issues that should be investigated further or of a permanent change in application usage that warrants VM sizing adjustments.
- Monthly Monitoring – A full-scale evaluation of an environment’s resource usage at a weekly time frame will reveal steady resource growth trends. By forecasting continued VM resource requirements based on the observed growth rate, virtualization administrators can assess whether they will have enough capacity available in the near future to avoid resource shortages. VM administrators can also factor in usage spikes seen from daily monitoring to assess what an environment’s resource usage "ceiling" is at peak periods.
To undertake the monitoring practices mentioned above, VM administrators should evaluate system metrics for CPU, memory, storage I/O, and storage space usage for all VMs and datastores in an environment. This metrics data is available within VMware’s vCenter and can be exported into Excel or viewed within vCenter in resource usage graphs set at different time scales. More details concerning the metrics to assess, and what values should alert of impending issues are provided in the recently released whitepaper VMware Performance Monitoring to Avoid Slow VMs.
Optimally, through a regular monitoring process, data centers should be able to foresee VM issues that would occur from resource shortages or capacity bottlenecks, and anticipate peak usage overflow for regularly recurring processes such as monthly in-depth virus scan. These insights allow virtualization administrators to preemptively rebalance resource usage loads or add more resources before a VMware performance problem occurs.
Automating the VM Monitoring Process
The steps described above can be time-consuming and must occur at regular intervals to effectively catch emerging VM performance issues. To automate this VM monitoring process, VKernel’s recently released vOPS Performance Analyzer, collects 20 performance metrics from vCenter every 30 minutes, monitoring resource usage at five minute granularity levels to detect utilization spikes. Additionally, Performance Analyzer allows virtualization administrators to create trending alarms to be warned when a particular resource’s usage has changed by a specified amount in a certain time frame. The end result is that after a 20 minute installation, the processes described above requires no recurring manual intervention. Instead the environment’s health can be assessed on a daily, weekly, and monthly basis from a summary report. Any notable issues can be set to generate immediate alarms.
Stay tuned for the next part of the Maintaining VM Performance blog series where we explore what to do if despite proactive monitoring efforts, a VM performance problem still occurs unexpectedly.