Autoscaling OpenStack by Cloud bursting

One of the essential characteristics of "Cloud" as specified in the NIST's definition of a cloud is "Rapid elasticity", which in our view implies being "unlimited". Our goal has been that enterprises that have private cloud should not be limited by the hardware they have been able to invest. For example, inspite of having a private cloud made possible by OpenStack, some online retail store/service might have infrequent and unpredictable spikes in traffic and investing on hardware to cater to these spikes might not be feasible.

For such enterprises to leverage their existing hardware using OpenStack and at the same time be able to automatically scale (autoscale the load on private cloud) by bursting into their public cloud of choice, a driver is being implemented. This will enable enterprises to save purchasing additional capacity which will be under utilised until a spike. This driver will act as the safety net of being able to fall on to a public cloud on demand. All this while continuing to be able to use the OpenStack native interfaces, Horizon or CLI or APIs. The OpenStack resources like keys, images, flavors and security groups and such are also mapped to the equivalents of the public cloud service provider.

The underlying principle is to define a public cloud to be just another hypervisor with unlimited resources.

This has been implemented keeping in mind that the investments might be limited, hence even a basic server configured as compute node with this driver, will integrate seamlessly into the existing ecosystem. A foreign availability zone is creating with this compute node. "nova-scheduler" which handles the requests will use the cloud bursting filter and costs/weights to load balance and burst to a public cloud when the internal cloud's utilisation has reached a specified limit.

Since public cloud comes with unlimited resources and keeping in mind that this driver is going to automatically scale to a public cloud without any human intervention, a quota limit has been implemented to specify the maximum resources that can be consumed on the public cloud.