An issue we have with the onapp autoscaling feature for virtual machines is that once a machine is auto-scaled up, it never scales itself back down to the default resource values assigned to the machine. A feature that allows for the virtual machine to scale back down to the provisioned specifications after a certain amount of time would be ideal along with a feature that allows us to set a maximum amount of resources that a machine can autoscale too, not just the 24h limit but rather a total autoscaled cap for the respective resource options.
At present with the current implementation, a virtual machine can perpetually continue to autoscale with no hard limit in place other than the per-24h scaling limit or the billing plan, which the billing plan can sometimes have huge amounts of resources that would be undesirable to assign to a single VM. Over the course of days or weeks it is possible for a customer machine to scale up by multiple GB of RAM or multiple orders of magnitude for cpu usage share. This is certainly not ideal and makes it difficult to provide autoscaling as a solution to our customers as there is no ceiling to what the customer can potentially be charged.
The short of it:
- Autoscale up with hard ceiling values instead of just 24h limits
- Autoscale down with the instances BASE resource values being the low level point
- Autoscale up period, after time has elapsed resources are returned to instances BASE values (appropriate for RAM/CPU)
- Correct monitis so that it takes free buffers/cache into account when determining free memory
Please sign in to leave a comment.