Scale on usage

A Usage-based Auto Scaling Group adds VMs when usage stays above the target you set, for example when CPU utilization is high, and removes them when usage stays below it. The group always keeps between Minimum VMs and Maximum VMs. Use this policy when your traffic is hard to predict. To change capacity at fixed times instead, see Scale on a schedule.

Before you begin

• An Auto Scaling Group, or the Create Auto Scaling Group form open. See Create an Auto Scaling Group.
• A usage level that one VM of your plan handles comfortably. You base the target on it.

Choose the policy and the capacity range

The Scaling section is on the Create Auto Scaling Group form and on Edit Auto Scaling Group. You can edit a group while it is running or stopped.

1. In Scaling, set Minimum VMs and Maximum VMs. They start at 2 and 5. The minimum can be 1 up to one less than the maximum, and the maximum can be at most 60.
2. Keep Policy on Usage-based, the default.

Left to itself, the group stays between Minimum VMs and Maximum VMs. A higher minimum absorbs a sudden spike while new VMs boot. The maximum caps what the group can cost.

Choose a metric and a target

Choose the metric to keep the group’s average around, then the target for it.

Metric	What the group watches	Target
CPU utilization	The share of the VMs’ vCPUs in use, over the last 2 minutes. This is the default.	A percentage, 10 to 99
Memory utilization	The memory in use divided by the memory given to the VMs.	A percentage, 10 to 99
CPU + Memory utilization	Both. The group adds VMs when either is over its target and removes them only when both are under.	Two percentages, 10 to 99 each
Network throughput	Bytes received plus bytes sent per second, averaged per VM, over the last 2 minutes. One MB/s is 1,000,000 bytes.	MB/s per VM, more than 0

The group averages the metric across all its VMs. The target starts at 60, and for CPU + Memory utilization the memory target starts at 70. For Network throughput, the target starts at 10 MB/s. Set the network target to what one VM handles comfortably, because the group adds capacity once the average goes above it.

Each metric has one target, and that one target controls both adding and removing VMs.

Choose how fast the group reacts

For a Usage-based group, Scaling behavior under Advanced Settings sets how long usage has to stay past the target before the group acts. Pick a Preset:

Preset	Use it for	Scale-out stabilization	Scale-in stabilization
Responsive	Sudden traffic spikes. It adds capacity fast.	60 seconds	300 seconds
Balanced (recommended)	A balance of stability and cost. This is the default.	150 seconds	600 seconds
Cost optimized	Fewer VMs for less of the time. It waits longer before changing.	300 seconds	1,200 seconds

Scale-out stabilization (seconds) is how long usage must stay above the target before the group adds VMs. It is at least 1. Scale-in stabilization (seconds) is how long usage must stay below the target before the group removes VMs. It is at least 120, so that a new VM can finish booting. Any other pair of numbers is a Custom preset.

Check what the group will do

Before you save, read How it will scale. It sums up your choices in plain words, for example “2 VMs to 5 VMs, targeting 60% CPU utilization”, and says what the group does at each end.

Save with Create Auto Scaling Group or Update Auto Scaling Group. The change takes effect at once on a running group.

Each time usage changes the group’s size, Scaling Activity records how many VMs were added or removed and why, such as “CPU utilization reached 72% (target 60%).” TIR also warns you when a group changes size back and forth often, six or more times in 24 hours with three or more switches between adding and removing, or reaches its maximum two or more times in 24 hours. Either is a sign to adjust the target, the range, or the scaling behavior.

To add capacity at set times on top of this policy, see Scale on a schedule. To add VMs right now, see Use Scale Now. To replace VMs that stop responding, see Set up health checks.

For how the group decides when to add and remove VMs, see How auto scaling works.