What is it?
In summary, it’s hot-spot detection for Nutanix Clusters running the AHV hypervisor with a unique feature that other hypervisors don’t have – it’s compute and storage aware. The functionality is enabled by default but can be disabled if required.
Because the AOS architecture (a) controls and (b) is aware of all I/O from the ground up the Controller Virtual Machines (CVMs) can make decisions based anomalies. So, VMs that peak in CPU and/or storage utilisation (> 85%) resulting in a “hot-spot” will force the cluster to identify the least used host using fair-weight queueing. A calculation is made to ascertain if it’s efficient to live-migrate the ‘trouble’ VM or the VMs around it – the least cost of time is always preferred.
Let’s be clear here too, it does not run a monitoring task every second. Not only is that approach entirely inefficient and pointless but it would burn valuable CPU cycles. Instead, AOS polls every 15 minutes and reviews its results and make decisions accordingly. In the event a rebalance occurs the subsequent poll will not happen for a further 30 minutes, after this, it’s back to 15 minutes. Note, there is no option to force a scan manually through the UI or command line.