Cloned from launchpad blueprint https://blueprints.launchpad.net/nova/+spec/multi-back-ends-for-nova-compute.
Description:
#1. Efficient node usage: Nova-compute is designed to configure with only one type of nova compute driver like kvm, vmware, citrix,etc. Here Hypervisors like kvm, the nova-compute is installed on the hypervisor itself.
So the CPU, memory load will be high as hypervisor is running on the nova-compute. But for the vmware hypervisors, nova-compute is installed on the separate node and its called proxy nova-compute. So the load on the proxy-compute node will be comparatively lesser and it will be more efficient if nova-compute node is allowed to configure multiple compute proxy driver similar to the cinder-volume multi-backend architecture. https://wiki.openstack.org/wiki/Cinder-multi-backend
#2. scalability: By enabling the multiple back-ends, it also helps to improve the scale-ability of the given compute proxy driver such as vmware vc driver.
Implementation Note
===============
It can be implemented by leveraging the cinder implementation where only one instance of cinder-volume is installed, but when the cinder-volume service is started, it spawns one cinder-volume process for every back-end configured in the cinder.conf. In the same approach,
nova.conf will be as follows:
[DEAFULT]
....
#back_ends will have comma separated set of compute drivers section
#Each compute driver could be same or different
back_ends = vcdriver1, vcdriver2, xen1
[vcdriver1]
...
[vcdriver2]
...
[xen1]
...
With these configuration in place, when nova-compute starts, it will spawn 3 nova-compute one for each of the configured driver.
Specification URL (additional information):
None